WO2024065988A1 - 设备监测方法和系统、电子设备、存储介质 - Google Patents

设备监测方法和系统、电子设备、存储介质 Download PDF

Info

Publication number
WO2024065988A1
WO2024065988A1 PCT/CN2022/133182 CN2022133182W WO2024065988A1 WO 2024065988 A1 WO2024065988 A1 WO 2024065988A1 CN 2022133182 W CN2022133182 W CN 2022133182W WO 2024065988 A1 WO2024065988 A1 WO 2024065988A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
target
analyzed
model
risk
Prior art date
Application number
PCT/CN2022/133182
Other languages
English (en)
French (fr)
Inventor
王宗文
王新梦
曹璞
Original Assignee
烟台杰瑞石油装备技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 烟台杰瑞石油装备技术有限公司 filed Critical 烟台杰瑞石油装备技术有限公司
Publication of WO2024065988A1 publication Critical patent/WO2024065988A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3065—Monitoring arrangements determined by the means or processing involved in reporting the monitored data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3089—Monitoring arrangements determined by the means or processing involved in sensing the monitored data, e.g. interfaces, connectors, sensors, probes, agents
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/32—Monitoring with visual or acoustical indication of the functioning of the machine
    • G06F11/324—Display of status information
    • G06F11/327—Alarm or error message display

Definitions

  • the present disclosure generally relates to the field of cloud computing technology, and more specifically to a device monitoring method and system, an electronic device, and a storage medium.
  • equipment fault identification and predictive maintenance can be performed through online monitoring systems.
  • the online monitoring system obtains equipment data collected by edge acquisition devices at a certain frequency, and performs equipment health status assessment and fault identification based on the equipment data.
  • the present disclosure provides a device monitoring method, which includes: acquiring current data to be analyzed of a target device according to a current data acquisition method, wherein the current data to be analyzed is data of a target data type collected from the target device; analyzing the current data to be analyzed using a target decision model to obtain a target monitoring result, wherein the target decision model is used to analyze the data of the target data type to obtain a monitoring result of the target monitoring type; determining a target data acquisition method corresponding to the target monitoring result, updating the current data acquisition method through the target data acquisition method, and obtaining an updated data acquisition method; acquiring the latest data to be analyzed according to the updated data acquisition method, and continuing to monitor the target device.
  • the present disclosure also provides a device monitoring system, which includes an acquisition terminal and an algorithm decision terminal: the acquisition terminal is configured to acquire data to be analyzed of a target data type from a target device; the algorithm decision terminal is configured to acquire current data to be analyzed from the acquisition terminal according to a current data acquisition method, wherein the current data to be analyzed is data of the target data type acquired by the acquisition terminal from the target device; the current data to be analyzed is analyzed using a target decision model to obtain a target monitoring result, wherein the target decision model is used to analyze data of the target data type to obtain a monitoring result of the target monitoring type; a target data acquisition method corresponding to the target monitoring result is determined, and the current data acquisition method is updated through the target data acquisition method to obtain an updated data acquisition method; the latest data to be analyzed is acquired from the acquisition terminal according to the updated data acquisition method, and the target device continues to be monitored.
  • a device monitoring system which includes an acquisition terminal and an algorithm decision terminal: the acquisition terminal is configured to acquire data to be analyzed of a target data
  • the present disclosure further provides a storage medium, which includes a stored program, and the method of the present disclosure is executed when the program is run.
  • the present disclosure further provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; the memory is used to store a computer program; and the processor is used to implement the method described in the present disclosure when executing the computer program.
  • the present disclosure provides a computer-readable storage medium, wherein the storage medium includes a stored program, wherein the method described in the present disclosure is executed when the program is run.
  • the present disclosure adopts a method of acquiring current data to be analyzed of a target device according to a current data acquisition method, wherein the current data to be analyzed is data of a target data type collected from the target device; the current data to be analyzed is analyzed using a target decision model to obtain a target monitoring result, wherein the target decision model is used to analyze data of the target data type to obtain a monitoring result of the target monitoring type; a target data acquisition method corresponding to the target monitoring result is determined, and the current data acquisition method is updated through the target data acquisition method to obtain an updated data acquisition method; the latest data to be analyzed is acquired according to the updated data acquisition method, and the target device is continued to be monitored, and the acquired device data is analyzed through the target decision model to obtain a real-time monitoring result of the target device, and the device data acquisition method is dynamically determined based on the real-time monitoring result of the target device, thereby achieving the purpose of automatically optimizing the data acquisition method according to the current actual situation of the target device, thereby solving
  • FIG1 is a schematic diagram of a hardware environment of a device monitoring method provided by an embodiment of the present disclosure
  • FIG2 is a flow chart of an optional device monitoring method provided by an embodiment of the disclosure.
  • FIG3 is a schematic diagram of the overall functions and communication framework of an optional equipment monitoring system provided by an embodiment of the present disclosure
  • FIG4 is a schematic diagram of a communication framework between an acquisition terminal and an algorithm decision system of an optional equipment monitoring system provided in an embodiment of the present disclosure
  • FIG5 is a schematic diagram of a communication framework between an algorithm decision system and an algorithm expert system of an optional equipment monitoring system provided in an embodiment of the present disclosure
  • FIG6 is a schematic diagram of an optional equipment monitoring system provided by an embodiment of the present disclosure.
  • FIG7 is a schematic diagram of another optional equipment monitoring system provided by an embodiment of the present disclosure.
  • FIG8 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure.
  • the present disclosure provides an embodiment of a method for monitoring equipment.
  • the device monitoring method of the present disclosure can be applied to a hardware environment composed of a terminal 101 and a server 103 as shown in FIG1 .
  • the server 103 is connected to the terminal 101 via a network, and can be used to provide device monitoring services for the terminal or a client installed on the terminal.
  • a database 105 can be set on the server or independently of the server to provide data storage services for the server 103.
  • the above-mentioned network includes but is not limited to: a wide area network, a metropolitan area network or a local area network, and the terminal 101 is not limited to a PC, a mobile phone, a tablet computer, etc.
  • the device monitoring method of the embodiment of the present disclosure can be executed by the server 103, or by the terminal 101, or by the server 103 and the terminal 101.
  • the terminal 101 executes the device monitoring method of the embodiment of the present disclosure, or it can be executed by the client installed thereon.
  • the following is an example of executing the device monitoring method of the present disclosure on the server.
  • FIG. 2 is a flow chart of an optional device monitoring method according to an embodiment of the present disclosure. As shown in FIG. 2 , the method may include the following steps:
  • Step S202 obtaining the current data to be analyzed of the target device according to the current data acquisition method, wherein the current data to be analyzed is data of the target data type collected from the target device;
  • Step S204 using the target decision model to analyze the current data to be analyzed to obtain the target monitoring result, wherein the target decision model is used to analyze the data of the target data type to obtain the monitoring result of the target monitoring type;
  • Step S206 determining a target data acquisition method corresponding to the target monitoring result, and updating the current data acquisition method by the target data acquisition method to obtain an updated data acquisition method
  • Step S208 obtaining the latest data to be analyzed according to the updated data acquisition method, and continuing to monitor the target device.
  • the current data to be analyzed of the target device is obtained according to the current data acquisition method, wherein the current data to be analyzed is data of the target data type collected from the target device; the current data to be analyzed is analyzed using the target decision model to obtain the target monitoring result, wherein the target decision model is used to analyze the data of the target data type to obtain the monitoring result of the target monitoring type; the target data acquisition method corresponding to the target monitoring result is determined, and the current data acquisition method is updated through the target data acquisition method to obtain an updated data acquisition method; the latest data to be analyzed is obtained according to the updated data acquisition method, and the target device is continued to be monitored, which can solve the technical problem in the related technology that the equipment monitoring system cannot automatically optimize the data acquisition method, thereby achieving the technical effect of improving the monitoring capability of the equipment monitoring system.
  • the equipment monitoring method disclosed in the present invention can be applied to a variety of intelligent decision-making production scenarios, including but not limited to: equipment fault identification prediction, equipment capacity and output prediction and equipment production scheduling, on-site parts life cycle and inventory prediction, etc.
  • the server obtains the current data to be analyzed of the target device according to the current data acquisition method, wherein the current data to be analyzed is data of the target data type collected by the target device;
  • the data to be analyzed is the data of the target data type collected from the target device. By analyzing the data to be analyzed, the monitoring results of the device status of the target device can be obtained.
  • the current data to be analyzed refers to the data that has not been obtained before the current moment and is used to analyze the device status of the target device.
  • the target data type is the data type corresponding to the purpose of equipment monitoring. For example, if the purpose of equipment monitoring is to identify and predict equipment faults, then the target data type is the data type used for equipment fault identification, such as equipment temperature data, equipment pressure data, equipment vibration data, equipment displacement data, etc.
  • the current data to be analyzed of the target device can be obtained from the acquisition terminal, which refers to a terminal with hardware sensors (including but not limited to: pressure, temperature, vibration, displacement, power, etc.) and data acquisition software (i.e., a software system that collects and uploads digital signals at a certain acquisition frequency).
  • the acquisition terminal refers to a terminal with hardware sensors (including but not limited to: pressure, temperature, vibration, displacement, power, etc.) and data acquisition software (i.e., a software system that collects and uploads digital signals at a certain acquisition frequency).
  • the acquisition terminal is used to directly collect data of the target data type from the target device.
  • the various sensors of the acquisition terminal are installed at monitoring points near important vulnerable parts of the target device. Through power supply, the real-time acquisition of various sensor data at each measuring point is realized, and the data is sent out through the network on its software end.
  • the data can be sent to the server, other terminals, cloud, etc.
  • the acquisition terminal can realize data acquisition through hardware sensors and digital signal analysis and upload through acquisition software. Whether the specific hardware and software are integrated or deployed separately is not limited.
  • the server obtains the current data to be analyzed of the target device according to the current data acquisition method, and may obtain the current data to be analyzed of the target device from the acquisition terminal according to the current data acquisition method.
  • the data acquisition method may include a first acquisition method and a second acquisition method.
  • the first acquisition method is to acquire data according to a first cycle
  • the second acquisition method is to acquire data according to a second cycle
  • the second cycle is shorter than the first cycle.
  • the first acquisition method can be a question-and-answer method, which is: sending an acquisition instruction to the acquisition terminal (the acquisition instruction can be sent at equal intervals according to the time period length T1), and the acquisition terminal responds to the acquisition instruction and sends the current data to be analyzed.
  • the characteristics of the question-and-answer method it can reduce the accumulation of raw data, perform data acquisition communication at equal time intervals, and the acquisition frequency is relatively low.
  • the second acquisition method can be a reporting method, in which the acquisition terminal actively reports the current data to be analyzed at a preset frequency.
  • the acquisition terminal can upload the data to be analyzed to a server, other terminals, the cloud, etc. at equal lengths and intervals, at high speed, and in real time through protocols such as mqtt, file ftp, tcp, and upd, according to a time period length T2, wherein T2 is less than T1.
  • T2 time period length
  • the first acquisition method when acquiring the data to be analyzed from the target device for the first time, can be used as the current acquisition method, that is, data acquisition is performed at a lower frequency. This is because when the target device is just being monitored, the target device is usually still in a normal state, and a lower frequency of data acquisition can already meet the monitoring needs.
  • time period for acquiring data can be determined according to actual needs. Different data acquisition methods can have different time periods.
  • Data transmission can be carried out through protocols including but not limited to Modbus-tcp, EtherNet/IP, 7S, scokect, http, etc.
  • the server uses the target decision model to analyze the current data to be analyzed to obtain the target monitoring result, wherein the target decision model is used to analyze the data of the target data type to obtain the monitoring result of the target monitoring type.
  • the type and form of the target decision model are not limited. It can be a judgment model based on preset rules, that is, a white box analysis model, or a prediction model based on a neural network, that is, a black box analysis model, or a combination of the two.
  • the target decision model can include multiple models, as well as decision rules for obtaining monitoring results based on the analysis results of multiple models.
  • the target decision model can be a model based on the fault identification rules of the target device, or a fault prediction model based on a neural network, or a combination of the two.
  • the white box analysis model includes but is not limited to the following three types: A. Calculate the mean square error (effective value), mean, absolute square value, variance, standard deviation, peak value, peak-to-peak value, maximum value, minimum value, waveform index, pulse index, margin, kurtosis, etc. corresponding to the original data of various types of signals with specified equal time intervals n or equal time intervals n corresponding to angular domains, and generate graded alarms for the timing fluctuations of the above indicators exceeding the threshold by setting custom thresholds; B.
  • sampling is performed according to equal time intervals n or equal time intervals n corresponding to angular domains, and all sample data are stacked to obtain the normal state of each measuring point of the equipment under different working conditions.
  • the signal distribution base space of the normal state is set, and the threshold or rule of exceeding the range of the basic space is set to identify the fault state of the data to be tested and give graded alarms.
  • the data range interval or threshold of the normal state of the equipment is set for the thermodynamic indicators such as temperature, pressure, displacement, etc. at the specified measuring point.
  • the data range or threshold of the normal state it is judged whether the signal data of the current equipment to be tested exceeds the range or the threshold and there is a fault risk, and graded alarms are given.
  • the normal working temperature of the plunger pump reducer is up to 110°C, and the maximum upward fluctuation range does not exceed 20°C.
  • the temperature range of the plunger pump reducer in the normal state is set to [110°C, 130°C], and a fault risk alarm is given according to whether the actual temperature exceeds this range.
  • Equal time interval n or equal time interval n corresponds to equal angle interval in angular domain:
  • a single time period corresponds to a single angle period, that is, within a single operation time period, the corresponding crankshaft rotates one circle, that is, 0-360°, so the single time period can correspond to a complete angle period;
  • the single time period corresponding to the reciprocating motion of the piston rod or the pull rod is the time for the crankshaft connected to the pull rod to rotate one circle, that is, 0-360°. Therefore, a single time period also corresponds to a single angle period. That is, all equipment unit time periods can be converted to single angle periods as a scale for analysis or evaluation.
  • variable speed it is necessary to intercept the time interval at different speeds according to the corresponding time points of the speed values in different time periods, that is, calculate the single cycle time length by the speed, and then truncate the time domain data according to the single cycle time length at different speeds, and divide the single cycle data of different time lengths at different speeds according to 0° ⁇ 360° equal angle intervals to obtain single cycle time domain data and corresponding angle domain data.
  • Stack all sample data For example, in each unit time period or unit angle period, the index corresponds to 0 to 99, with a total of 100 points.
  • the amplitude corresponding to each index of n single time period or single angle period data 0 to 99 is averaged, maximum value, 3/4 quantile, upper envelope value, etc., and the statistical indicators under each index of 0 to 99 are connected to obtain the signal distribution reference space.
  • black box analysis models include but are not limited to the following two types:
  • the network structure of this type of model is mainly based on self-supervised or unsupervised neural network, including but not limited to AE autoencoder, SAE and other autoencoder variants, in which the input data enters the encoder for dimensionality reduction feature extraction, the encoder result is input to the decoder to restore the original data, and the difference between the decoding result and the input data is calculated.
  • the input and output data error threshold the current fault risk level of the equipment is predicted and graded warning is given
  • the network structure of this type of model is mainly based on self-supervised or unsupervised neural network, including but not limited to AE autoencoder, SAE and other autoencoder variants.
  • the network structure is mainly based on supervised neural networks, including but not limited to supervised neural network models such as CNN, RNN, LSTM and their variants, where the input data include but not limited to: time domain and frequency domain data sampled at equal time intervals or equal angle intervals of various signals, or various characteristic indicator data corresponding to the time and frequency domain (including but not limited to mean square error (effective value), mean, absolute square value, variance, standard deviation, peak value, peak-to-peak value, maximum value, minimum value, waveform index, pulse index, margin, kurtosis, etc.), time-frequency diagram or time-frequency matrix, etc.; the output is the current risk level label of the equipment, and the label type corresponds to the equipment risk warning level, namely: normal equipment status, medium risk of equipment failure, and high risk of equipment failure.
  • the target decision model can also be a combination of a white-box analysis model and a black-box analysis model: the current data to be analyzed is analyzed by the white-box analysis model to obtain a first analysis result, wherein the white-box analysis model is a model based on the fault identification rules of the target device, and the target decision model includes the white-box analysis model; the current data to be analyzed is analyzed by the black-box analysis model to obtain a second analysis result, wherein the black-box analysis model is a fault prediction model based on a neural network, and the target decision model includes the black-box analysis model; based on the first analysis result and the second analysis result, as well as the target decision rule, a target monitoring result corresponding to the current data to be analyzed is obtained, wherein the target decision model includes the target decision rule, and the target decision rule is used to determine the monitoring result according to the analysis result of the data by each analysis model in the target decision model, and the monitoring result is used to indicate the failure risk of the target device.
  • Target monitoring types include but are not limited to the following: equipment failure risk monitoring, equipment capacity and output forecasting, equipment production scheduling, on-site spare parts life cycle and inventory forecasting.
  • the form of the monitoring result is not limited, and it can be a label result, a numerical result, etc.
  • the monitoring result can be a label indicating the failure risk level of the target equipment, such as "equipment normal”, “equipment low risk”, “equipment medium risk”, “equipment high risk”; the monitoring result can be a numerical value indicating the failure risk level of the target equipment, and the higher the numerical value, the higher the failure risk.
  • the server determines a target data acquisition method corresponding to the target monitoring result, updates the current data acquisition method by the target data acquisition method, and obtains an updated data acquisition method.
  • different data acquisition methods can be used for different situations of the target equipment to achieve better monitoring results. For example, in the scenario of equipment failure risk monitoring, if the current failure risk level of the target equipment is medium or high risk, it is necessary to acquire data at a higher data acquisition frequency to accurately and real-time capture the equipment failure characteristics in the data to obtain more accurate monitoring results; if the current failure risk level of the target equipment is zero risk or low risk, it is not necessary to acquire data at a higher data acquisition frequency. If data is still acquired at a higher data acquisition frequency, it may lead to problems such as data accumulation and waste of computing power.
  • the server can learn the current situation of the target device based on the target monitoring result, determine the data acquisition method that matches the current situation, and acquire data according to the updated data acquisition method. For example, if the target monitoring result indicates that the failure risk of the target device is high, the current acquisition method needs to be updated from the first acquisition method with a longer period to the second acquisition method with a shorter period.
  • the server obtains the latest data to be analyzed according to the updated data acquisition method and continues to monitor the target device.
  • Monitoring the target device refers to the process of obtaining the data of the target device and analyzing it to obtain the monitoring results, as well as determining the data acquisition method based on the monitoring results.
  • the monitoring of the device is long-term monitoring and real-time monitoring, that is, the latest data to be analyzed is continuously obtained during the monitoring process, and the monitoring results corresponding to the data to be analyzed are continuously analyzed.
  • the data acquisition method suitable for the latest situation of the target device is continuously determined based on the latest monitoring results.
  • the latest data to be analyzed is obtained according to the updated data acquisition method, and the target decision model is used to analyze the latest data to be analyzed to obtain the latest monitoring results, and the latest data acquisition method is determined based on the latest monitoring results.
  • the target monitoring type is fault risk monitoring.
  • step S206 a target data acquisition method corresponding to the target monitoring result is determined, and the following steps are also included:
  • Step S31 when the current data acquisition mode is the first acquisition mode of acquiring data according to the first cycle, and the target monitoring result indicates that the failure risk of the target device is less than or equal to the risk lower limit, determining that the target data acquisition mode is the first acquisition mode;
  • Step S32 when the current data acquisition method is the first acquisition method and the target monitoring result indicates that the failure risk of the target device is greater than the risk lower limit, determine that the target data acquisition method is the second acquisition method of acquiring data according to the second cycle, wherein the second cycle is shorter than the first cycle.
  • the first acquisition method can be a question-and-answer method, which is: sending an acquisition instruction to the acquisition terminal (the acquisition instruction can be sent at equal intervals according to the time period length T1), and the acquisition terminal responds to the acquisition instruction and sends the current data to be analyzed.
  • the characteristics of the question-and-answer method it can reduce the accumulation of raw data, perform data acquisition communication at equal time intervals, and the acquisition frequency is relatively low.
  • the second acquisition method can be a reporting method, which is: the acquisition terminal actively reports the current data to be analyzed at a preset frequency.
  • the acquisition terminal can upload the data to be analyzed to the server, other terminals, the cloud, etc. at equal lengths and intervals, at high speed, and in real time through protocols such as mqtt, file ftp, tcp, and upd, according to the time period length T2, where T2 is less than T1.
  • the characteristics of the reporting method data is collected in real time with zero delay and high frequency to ensure accurate and real-time capture of equipment failure characteristics in the data.
  • the first acquisition method can be used as the current acquisition method, that is, data acquisition is performed at a lower frequency. This is because when the target device is just being monitored, the target device is usually still in a normal state, and a lower frequency of data acquisition can already meet the monitoring needs.
  • the target monitoring result indicates that the failure risk of the target device is greater than the risk lower limit
  • the target device is not abnormal at present, and data can be acquired at the original lower frequency without changing the data acquisition method. That is, the target data acquisition method is determined to be the first acquisition method of acquiring data according to the first period.
  • the target monitoring result indicates that the failure risk of the target device is greater than the risk lower limit
  • the target device is currently beginning to show abnormalities and may have a failure
  • high-frequency data acquisition is required to ensure accurate and real-time capture of the device failure characteristics in the data, that is, the target data acquisition method is determined to be the second acquisition method of acquiring data according to the second period.
  • a data acquisition method that acquires data at a lower frequency is adopted.
  • a data acquisition method that acquires data at a higher frequency is adopted instead.
  • step S204 after the target decision model is used to analyze the current data to be analyzed and the target monitoring result is obtained, the method further includes the following steps:
  • Step S321 when the current data acquisition method is the first acquisition method and the target monitoring result indicates that the failure risk of the target device is greater than the risk lower limit, a specified risk event is generated, wherein the specified risk event is used to indicate that the failure risk of the target device changes from less than or equal to the risk lower limit to greater than the risk lower limit.
  • the target monitoring result indicates that the failure risk of the target device is greater than the risk lower limit
  • the target device is currently beginning to show abnormalities and may have a failure.
  • a risk event needs to be generated and the relevant data of the risk event needs to be recorded in the database table so that further risk judgment and risk control strategies can be taken later.
  • step 206 determining a target data acquisition method corresponding to the target monitoring result, further includes the following steps:
  • Step S322 determine that the target data acquisition method corresponding to the target monitoring result is the second acquisition method, and update the current data acquisition method from the first acquisition method to the second acquisition method through the target data acquisition method.
  • step 206 is executed in the manner of step S322.
  • step 208 obtaining the latest data to be analyzed according to the updated data acquisition method and continuing to monitor the target device, also includes the following steps:
  • Step S323 Acquire the latest data to be analyzed according to the second acquisition method and continue to monitor the target device.
  • step 208 is executed in the manner of step S323.
  • step 323, obtaining the latest data to be analyzed according to the second acquisition method and continuing to monitor the target device also includes executing the following loop steps:
  • Step S3232 using the target decision model to analyze the latest data to be analyzed to obtain the latest monitoring results, wherein the target decision model is used to analyze data of the target data type to obtain the monitoring results of the target monitoring type.
  • the step of determining the data acquisition method is no longer required in the device monitoring process, because an abnormality has already occurred in the target device at this time. Even if the subsequent monitoring results indicate that the target device is normal, the risk of failure of the target device cannot be ruled out. Therefore, it is necessary to continue to acquire data according to the second acquisition method.
  • the data acquisition method can be changed from the second acquisition method to the first acquisition method, and the complete device monitoring steps can be continued in a loop.
  • the latest data to be analyzed is continuously acquired, and the monitoring results corresponding to the data to be analyzed are continuously analyzed.
  • the data acquisition method suitable for the latest situation of the target device is continuously determined based on the latest monitoring results, and data is acquired in the subsequent monitoring process according to the latest changed data acquisition method.
  • the method further includes the following steps:
  • Step S41 obtaining the latest monitoring result corresponding to each latest data to be analyzed obtained within a specified time period, wherein the latest monitoring result corresponding to each latest data to be analyzed is obtained by analyzing the latest data to be analyzed using a target decision model;
  • Step S42 Determine a risk control strategy corresponding to a specified risk event according to the target monitoring result and the latest monitoring result.
  • the acquisition terminal uses the reporting method as the updated data acquisition method.
  • the acquisition terminal reports the latest data to be analyzed to the server at time intervals such as the cycle length T.
  • the server analyzes the latest data to be analyzed each time it reports to obtain the latest monitoring results corresponding to each report.
  • the server obtains the latest monitoring results corresponding to the latest data to be analyzed each time within the specified time period t, and obtains t/T time series of continuous monitoring results, where T is the length of the data acquisition cycle.
  • the classification rules of monitoring results and the determination method of risk control strategy rules are as follows:
  • step S41 after obtaining the latest monitoring result corresponding to each latest data to be analyzed obtained within a specified time period, the method further includes:
  • Step S43 using the target monitoring result and the latest monitoring result as the algorithm evaluation result of the specified risk event, and storing all the target data to be analyzed corresponding to the target monitoring result and the latest monitoring result as the original device data of the specified risk event in the historical risk event library, wherein the target data to be analyzed includes the current data to be analyzed and the latest data to be analyzed; and
  • Step S44 obtaining a designated evaluation result obtained by evaluating a designated risk event by the target object, and storing the designated evaluation result as an actual evaluation result of the designated risk event in a historical risk event library, wherein the actual evaluation result is used to indicate the actual identified failure risk of the corresponding risk event.
  • the server can obtain the actual evaluation results of the risk events from the expert evaluation interface.
  • the actual evaluation results are manual evaluations obtained by professionals after actual maintenance and professional analysis of the target equipment.
  • the evaluation mode is not limited. It can be a label selection mode or a scoring mode.
  • the evaluation mode is a three-value single choice, namely "definitely a high-risk failure", “definitely a medium-risk failure” or "definitely the equipment is normal".
  • the server will save the original device data, algorithm evaluation results, and actual evaluation results of each risk event to the corresponding database table in the background to generate a historical risk event record table for subsequent accuracy evaluation and update of the target decision model.
  • the method further comprises the steps of:
  • Step S51 obtaining a plurality of historical event data from a historical risk event library, wherein each historical event data uniquely corresponds to a historical risk event, and the historical event data corresponding to each historical risk event includes original device data corresponding to each historical risk event, an algorithm evaluation result, and an actual evaluation result;
  • Step S52 determining the current accuracy of the target decision model according to the algorithm evaluation results and the actual evaluation results in each historical event data
  • Step S53 when the current accuracy of the target decision model is lower than a preset threshold, the target decision model is updated according to a plurality of historical event data to obtain an updated target decision model.
  • the current accuracy of the target decision model (the number of historical risk events whose algorithm evaluation results are consistent with the actual evaluation results) / the total number of historical risk events.
  • the target decision model accuracy threshold ⁇ can be set to determine whether the current model prediction accuracy meets the requirements. If the target decision model prediction accuracy is less than ⁇ , the target decision model can be adjusted and updated.
  • step S53 updating the target decision model according to the plurality of historical event data to obtain an updated target decision model, further comprises the following steps:
  • Step S531 for each historical risk event corresponding to the historical event data, the algorithm evaluation result in the historical event data is corrected according to the actual evaluation result in the historical event data to obtain a corrected evaluation result of the historical risk event;
  • Step S532 based on the original equipment data, the revised evaluation results and the preset accuracy corresponding to each historical risk event, the black box analysis model in the target decision model is retrained to obtain an updated black box analysis model, wherein the black box analysis model is based on a neural network fault prediction model, and is used to analyze the data to be analyzed to obtain a black box analysis result, and the accuracy of the black box analysis result obtained by the updated black box analysis model for analyzing the data to be analyzed is not less than the preset accuracy;
  • Step S533 based on the original device data corresponding to each historical risk event, the revised evaluation result and the specified accuracy, adjust the judgment threshold of the fault identification rule in the white-box analysis model to obtain an updated white-box analysis model, wherein the white-box analysis model is a model based on the fault identification rule in the target decision model, and is used to analyze the data to be analyzed according to the fault identification rule to obtain a white-box analysis result, and the accuracy of the white-box analysis result obtained by the updated white-box analysis model for analyzing the data to be analyzed is not less than the specified accuracy; and
  • Step S534 based on the target decision rules in the target decision model, the updated black box analysis model and the updated white box analysis model, an updated target decision model is obtained, wherein the target decision rules are used to determine the monitoring results according to the analysis results of the data by each analysis model in the target decision model.
  • the update of the white-box analysis model can be done with the help of manual experience analysis by domain experts to determine the root cause of the white-box analysis model's misjudgment, and the threshold of the white-box analysis model can be appropriately adjusted and corrected based on expert experience, so that the prediction accuracy of the corrected white-box analysis model is improved to the specified accuracy.
  • step S204 using the target decision model to analyze the current data to be analyzed to obtain the target monitoring result, also includes the following steps:
  • Step S21 analyzing the current data to be analyzed by a white box analysis model to obtain a first analysis result, wherein the white box analysis model is a model based on a fault identification rule of a target device, and the target decision model includes the white box analysis model;
  • Step S22 analyzing the current data to be analyzed by a black box analysis model to obtain a second analysis result, wherein the black box analysis model is a fault prediction model based on a neural network, and the target decision model includes the black box analysis model; and
  • Step S23 based on the first analysis result and the second analysis result, and the target decision rule, obtain the target monitoring result corresponding to the current data to be analyzed, wherein the target decision model includes the target decision rule, and the target decision rule is used to determine the monitoring result according to the analysis result of the data by each analysis model in the target decision model, and the monitoring result is used to indicate the failure risk of the target equipment.
  • dual-track hybrid fault identification is performed based on the white-box analysis model and the black-box analysis model, and the target decision rule is as follows:
  • the accuracy of monitoring results can be improved, the failure risk of target equipment can be identified more sensitively, and the monitoring capability of the equipment monitoring system can be enhanced.
  • the algorithm decision system sets up a series of sensors to collect equipment data in real time, establishes a decision algorithm model for the equipment based on the historical data of equipment operation, and then uses this algorithm model to calculate the equipment data in real time, so as to evaluate the health status of the equipment and identify faults.
  • the online monitoring system generally includes three modules: edge acquisition equipment, edge algorithm decision system, and cloud algorithm expert system.
  • the three modules are executed independently and have no connection.
  • the acquisition frequency and transmission mode of the edge acquisition equipment are immutable during operation once they are set. If the acquisition frequency is set too high, it will cause a waste of memory computing power, and if it is set too low, it will be unfavorable for data analysis and fault diagnosis;
  • the edge algorithm decision system integrates a large number of decision algorithms for real-time decision-making, and after the algorithm is integrated, it is almost no longer upgraded or difficult to upgrade;
  • the algorithm in the cloud algorithm expert system is generally refined after analyzing a large amount of data, but after the algorithm is refined, it is no longer optimized and upgraded or cannot be upgraded due to lack of fault data, resulting in the algorithm only being suitable for a certain device and unable to upgrade adaptively.
  • the cloud-edge combined algorithm closed-loop decision system architecture method and a system for hierarchical decision-making to identify the risk level of equipment and automatic closed-loop upgrade and optimization of decision algorithms provided in this embodiment can link three modules, cloud-edge collaborative work, dynamically collect data, and automatically upgrade decision algorithms.
  • This embodiment can be applied to a variety of intelligent decision-making production scenarios, including but not limited to: equipment fault identification prediction, equipment capacity and output prediction and equipment production scheduling, on-site parts life cycle and inventory prediction, etc.
  • This embodiment mainly uses the equipment fault identification prediction scenario as an example to illustrate the method.
  • the system includes three modules: edge acquisition terminal, edge algorithm decision system, and cloud algorithm expert system.
  • Figure 3 is a schematic diagram of the overall function and communication framework of an optional equipment monitoring system in an embodiment of the present disclosure.
  • the edge system When the equipment is in normal state, the edge system communicates with the acquisition terminal through question-and-answer mode to obtain and store the original sensor signal data; the edge system deploys white-box rule models and black-box deep neural network models to perform dual-track hybrid fault identification and graded warning; the edge system switches the data communication mode with the terminal according to the fault identification and warning level; the edge system sends graded control instructions based on real-time and continuous diagnosis results; the edge PC system and app set up a manual evaluation entrance for the prediction accuracy of the fault identification model, and perform graded evaluation on the prediction accuracy of the risk event model; the edge system transmits historical fault warning events, the original data of sensors involved in the warning events, and the manual evaluation results of the warning accuracy of the event back to the cloud algorithm expert system through the network, and the algorithm expert system retrains and upgrades the model based on the historical accuracy of the algorithm and the original data of the warning events; after the cloud algorithm expert system completes the model upgrade iteration, the algorithm model is remotely transmitted to the edge system through the network for model replacement it
  • the equipment data acquisition terminal described in this embodiment refers to a terminal system with hardware sensors (including but not limited to: pressure, temperature, vibration, displacement, power, etc.) and data acquisition software (i.e., a software system that collects and uploads digital signals at a certain acquisition frequency).
  • Various sensors are installed at monitoring points near important vulnerable parts of the target equipment, and are powered by power to achieve real-time acquisition of various sensor data at each measuring point, and the data is sent out through the network at its software end.
  • the edge algorithm decision system described in this embodiment refers to a software system that uses white box and black box algorithm model services as the main body to identify the current state of the equipment failure risk.
  • the input of the system is the original data of various acquisition signals obtained from the equipment data acquisition terminal, and the output is the failure risk level of the equipment component of the measurement point corresponding to the current equipment acquisition signal.
  • the data acquisition terminal and the algorithm decision system communicate through question-and-answer mode.
  • the algorithm decision system actively obtains the latest data from the acquisition terminal at time intervals.
  • Data transmission can be through but not limited to: Modbus-tcp, EtherNet/IP, 7S, scokect, http and other protocols.
  • the above algorithm decision system mainly includes two parts of model services, namely white box and black box algorithm model service programs.
  • the white-box algorithm model service refers to a type of fault identification rule algorithm model.
  • the rule algorithm model refers to, but is not limited to, the following three types:
  • sampling is performed at equal time intervals n or equal angle intervals in the angular domain corresponding to equal time intervals n, and all sample data are stacked to obtain the normal state signal distribution reference space under different working conditions at each measuring point of the equipment, and the threshold or rule of exceeding the range of the basic space is set to identify the fault state of the data to be measured and give graded alarms;
  • thermodynamic indicators such as temperature, pressure, displacement, etc.
  • the data range or threshold of the normal state determine whether the signal data of the current equipment under test exceeds the range or threshold and there is a risk of failure, and give graded alarms;
  • the black box algorithm model service refers to a type of fault identification model based on deep neural networks.
  • This type of neural network algorithm model includes but is not limited to the following two types:
  • A. Equipment fault identification model based on self-supervised or unsupervised neural network The network structure of this type of model is mainly based on self-supervised or unsupervised neural network, including but not limited to AE autoencoder, SAE and other autoencoder variants.
  • the input data enters the encoder for dimensionality reduction feature extraction, and the encoder result is input into the decoder to restore the original data.
  • the difference between the decoding result and the input data is calculated, and the error threshold of the input and output data is set to predict the current fault risk level of the equipment and provide graded warning;
  • the main network structure of this type of model is supervised neural network, including but not limited to supervised neural network models such as CNN, RNN, LSTM and their variants.
  • the input data includes but is not limited to: time domain and frequency domain data sampled at equal time intervals or equal angle intervals of various signals, or various characteristic index data corresponding to the time and frequency domain (including but not limited to mean square error (effective value), mean, absolute square value, variance, standard deviation, peak value, peak-to-peak value, maximum value, minimum value, waveform index, pulse index, margin, kurtosis, etc.), time-frequency diagram or time-frequency matrix, etc.;
  • the output is the current risk level label of the equipment, and the label type corresponds to the equipment risk warning level, namely: normal equipment status, medium risk of equipment failure, and high risk of equipment failure.
  • the white-box model and the black-box model are combined according to the following rules to generate a dual-track identification result of the current risk level of the device.
  • Fig. 4 is a schematic diagram of the communication framework between the acquisition terminal and the algorithm decision system of an optional equipment monitoring system according to an embodiment of the present disclosure.
  • the algorithm decision system performs dual-track hybrid fault identification based on white boxes and black boxes, and outputs the current risk level of the equipment.
  • the result data will be fed back to the acquisition terminal in real time through communication protocols such as scokect, http, Modbus-tcp, EtherNet/IP, 7S, etc.
  • the acquisition terminal will trigger the data communication mode conversion instruction, and convert the question-and-answer communication mode to an active reporting mode.
  • the acquisition terminal will send the corresponding medium and high risk measurement point original signal data to the algorithm decision system through protocols such as mqtt, file ftp, tcp, upd, etc., according to the time period length T, equal length and interval, high speed, and real-time.
  • the acquisition terminal will send the signal time domain data to the algorithm decision system by actively reporting.
  • the algorithm decision system will create a table to store the original data of the risk event in the database, and perform dual-track fault identification and prediction for the real-time data with a cycle length of T for a continuous period of t, and output t/T time series continuous prediction results.
  • the algorithm system will perform hierarchical decision control based on the continuously output dual-track fault identification and prediction results, that is, different levels of decision control commands will be sent, so as to achieve the purpose of active decision-making on equipment operation and maintenance. Control commands are mainly sent in the form of text messages, mobile APP notifications, PC pop-ups, large-screen monitoring pop-ups, etc.
  • the main decision control levels and grading rules are as follows:
  • the algorithm decision system will set up an expert manual evaluation entrance for the prediction accuracy of the fault identification model on the PC side, mobile app, etc. Every time a risk event occurs, the on-site staff needs to manually confirm whether the risk notification or control instruction is executed. And after this risk event, combined with the actual maintenance and expert manual evaluation results, the algorithm fault identification prediction accuracy is evaluated.
  • the evaluation mode is a three-value single choice, that is, a high-risk fault, a medium-risk fault, or a normal equipment.
  • the algorithm decision system will save this risk event and the expert manual evaluation results to the corresponding database table in the background to generate a historical risk event record table.
  • the edge algorithm decision system regularly transmits the historical risk event record table and the original measurement point signal data corresponding to the historical risk events to the cloud algorithm expert system through message middleware (such as emqx, rabbitmq, kafka) or http interface, file ftp, etc., hereinafter referred to as the algorithm expert system.
  • the algorithm expert system is a support system for algorithm upgrade iteration and push.
  • the algorithm expert system backend database stores the historical risk event record table and the original measurement point signal data corresponding to the historical risk events from the algorithm decision system. By comparing the consistency of the risk level labels of historical risk events and the manual evaluation options, the accuracy of the algorithm model is obtained, that is:
  • Algorithm model accuracy (the number of risk events with the same risk level label and expert manual evaluation) / the total number of historical risk events
  • the algorithm model accuracy threshold ⁇ it is determined whether the current model prediction accuracy meets the requirements. If the model prediction accuracy is less than ⁇ , the algorithm expert system can provide a model retraining function. The process is as follows:
  • the source of risk level labels based on historical risk events is the signal data of the measuring point.
  • manual experience analysis is performed to determine the root cause of the white-box model's misjudgment.
  • the threshold of the white-box model is appropriately adjusted and corrected based on expert experience, so that the prediction accuracy of the corrected white-box model is improved by more than ⁇ , and the threshold corresponding to the corrected white-box model is output;
  • the algorithm expert system uploads the retrained black box model and the corrected Lily white box model threshold to the edge algorithm decision system through remote network transmission, and completes the model update deployment.
  • Figure 5 is a schematic diagram of the communication framework between the algorithm decision system and the algorithm expert system of an optional equipment monitoring system according to an embodiment of the present disclosure. Based on the above method, the whole process of cloud-edge combined equipment fault identification algorithm model decision and model closed-loop correction can be realized. Similarly, this method can also be used in other production scenarios, including but not limited to: equipment fault identification prediction, production equipment capacity and output prediction, on-site spare parts life cycle calculation and inventory prediction, etc.
  • a cloud-edge combined decision-making system is implemented, which uses an algorithm model as the driving force and can achieve full-link closed-loop upgrade and iteration. It is suitable for many industrial intelligent production application scenarios.
  • a data transmission method that implements hierarchical and graded command-based communication switching between the edge algorithm decision system and the signal acquisition terminal increases the acquisition transmission rate when the device is at high risk to ensure the validity of the data. When the device is at low risk, the rate is reduced to save memory computing power and improve system stability.
  • the closed-loop iterative upgrade and deployment application of the algorithm model based on the accuracy of the prediction results is realized between the cloud algorithm expert system and the edge algorithm decision system.
  • the algorithm is self-feedback, self-adaptive, and self-learning.
  • the combination of cloud and edge ensures that the fault data is transmitted to the cloud to the maximum extent, providing data for algorithm correction and optimization.
  • the algorithm is returned to the edge to form a closed-loop algorithm upgrade. Different equipment and different working conditions can be upgraded to the most suitable algorithm through this closed-loop method.
  • the edge algorithm decision system implements dual-track hybrid decision-making hierarchical control of rule-based white box algorithm model and neural network-based black box algorithm model.
  • the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method.
  • the technical solution of the present disclosure, or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, a disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present disclosure.
  • FIG6 is a schematic diagram of an optional device monitoring system 600 according to an embodiment of the present disclosure. As shown in FIG6 , the system 600 may include:
  • the acquisition terminal 22 is configured to acquire the target data type to be analyzed from the target device;
  • the algorithm decision terminal 24 is configured to obtain the current data to be analyzed from the acquisition terminal according to the current data acquisition method, wherein the current data to be analyzed is data of the target data type acquired by the acquisition terminal from the target device; analyze the current data to be analyzed using the target decision model to obtain the target monitoring result, wherein the target decision model is used to analyze the data of the target data type to obtain the monitoring result of the target monitoring type; determine the target data acquisition method corresponding to the target monitoring result, update the current data acquisition method through the target data acquisition method, and obtain an updated data acquisition method; obtain the latest data to be analyzed from the acquisition terminal according to the updated data acquisition method, and continue to monitor the target device.
  • the algorithm decision terminal 24 in this embodiment can be used to execute steps S202, S204, S206, and S208 in the embodiment of the present disclosure
  • the collection terminal 22 in this embodiment can be used to collect the data to be analyzed in steps S202 and S208 in the embodiment of the present disclosure.
  • acquisition terminal 22 and algorithm decision terminal 24 as part of the system can run in the hardware environment shown in FIG. 1 , and can be implemented by software or hardware.
  • the above-mentioned acquisition terminal 22 and algorithm decision terminal 24 analyze the acquired equipment data through the target decision model to obtain the real-time monitoring results of the target equipment, and dynamically determine the equipment data acquisition method based on the real-time monitoring results of the target equipment, thereby achieving the purpose of automatically optimizing the data acquisition method according to the current actual situation of the target equipment, thereby solving the technical problem in the related technology that the equipment monitoring system cannot automatically optimize the data acquisition method, and achieving the technical effect of improving the monitoring capability of the equipment monitoring system.
  • the algorithm decision terminal 24 may include: a data acquisition module, configured to acquire current data to be analyzed from the acquisition terminal according to the current data acquisition method, wherein the current data to be analyzed is data of the target data type acquired by the acquisition terminal from the target device; an algorithm decision module, configured to analyze the current data to be analyzed using a target decision model to obtain a target monitoring result, wherein the target decision model is used to analyze the data of the target data type to obtain a monitoring result of the target monitoring type; determine a target data acquisition method corresponding to the target monitoring result, and an acquisition method update module, configured to update the current data acquisition method through the target data acquisition method to obtain an updated data acquisition method.
  • a data acquisition module configured to acquire current data to be analyzed from the acquisition terminal according to the current data acquisition method, wherein the current data to be analyzed is data of the target data type acquired by the acquisition terminal from the target device
  • an algorithm decision module configured to analyze the current data to be analyzed using a target decision model to obtain a target monitoring result, wherein the target decision model is used
  • the data acquisition module in this embodiment can be used to execute step S202 in the embodiment of the present disclosure
  • the algorithm decision module in this embodiment can be used to execute step S204 in the embodiment of the present disclosure
  • the acquisition method update module in this embodiment can be used to execute step S206 in the embodiment of the present disclosure
  • the data acquisition module, algorithm decision module, and acquisition method update module in this embodiment can be used to jointly execute step S208 in the embodiment of the present disclosure.
  • FIG. 7 is a schematic diagram of another optional equipment monitoring system 700 according to an embodiment of the present disclosure.
  • the system 700 may include a collection terminal 22 , an algorithm decision terminal 24 , and an algorithm expert terminal 26 .
  • the acquisition method update module is further configured to: when the current data acquisition method is the first acquisition method of acquiring data according to the first cycle, and the target monitoring result indicates that the failure risk of the target device is less than or equal to the risk lower limit, determine that the target data acquisition method is the first acquisition method; when the current data acquisition method is the first acquisition method, and the target monitoring result indicates that the failure risk of the target device is greater than the risk lower limit, determine that the target data acquisition method is the second acquisition method of acquiring data according to the second cycle, wherein the second cycle is shorter than the first cycle.
  • the algorithm decision end 24 may include an event data module configured to: when the current data acquisition method is the first acquisition method and the target monitoring results indicate that the failure risk of the target device is greater than the risk lower limit, generate a specified risk event, wherein the specified risk event is used to indicate that the failure risk of the target device changes from less than or equal to the risk lower limit to greater than the risk lower limit.
  • the event data module is further configured to: when the current acquisition method is the first acquisition method and the updated data acquisition method is the second acquisition method, after acquiring the latest data to be analyzed according to the updated data acquisition method, acquire the latest monitoring result corresponding to each latest data to be analyzed acquired within a specified time period, wherein the latest monitoring result corresponding to each latest data to be analyzed is obtained by analyzing the latest data to be analyzed using the target decision model.
  • the algorithm decision end 24 may further include a risk strategy determination module configured to determine a risk control strategy corresponding to a specified risk event according to the target monitoring results and the latest monitoring results.
  • the event data module is further configured to: use the target monitoring results and the latest monitoring results as algorithm evaluation results of the specified risk event, and use all target data to be analyzed corresponding to the target monitoring results and the latest monitoring results as original equipment data of the specified risk event, and store them in the historical risk event library, wherein the target data to be analyzed includes current data to be analyzed and the latest data to be analyzed; obtain the specified evaluation result obtained by the target object to evaluate the specified risk event, and store the specified evaluation result as the actual evaluation result of the specified risk event in the historical risk event library, wherein the actual evaluation result is used to indicate the actual identified fault risk of the corresponding risk event.
  • the system may further include an algorithm expert terminal 26, including: a historical data acquisition module, used to acquire multiple historical event data from a historical risk event library, wherein each historical event data uniquely corresponds to a historical risk event, and the historical event data corresponding to each historical risk event includes the original equipment data corresponding to each historical risk event, the algorithm evaluation result, and the actual evaluation result; a model evaluation module, used to determine the current accuracy of the target decision model according to the algorithm evaluation result and the actual evaluation result in each historical event data; a model update module, used to update the target decision model according to multiple historical event data when the current accuracy of the target decision model is lower than a preset threshold, so as to obtain an updated target decision model.
  • a historical data acquisition module used to acquire multiple historical event data from a historical risk event library, wherein each historical event data uniquely corresponds to a historical risk event, and the historical event data corresponding to each historical risk event includes the original equipment data corresponding to each historical risk event, the algorithm evaluation result, and the actual evaluation result
  • a model evaluation module used to determine the current accuracy
  • the model updating module is further configured to: for each historical risk event corresponding to the historical event data, according to the actual evaluation result in the historical event data, correct the algorithm evaluation result in the historical event data to obtain a corrected evaluation result of the historical risk event; based on the original equipment data corresponding to each historical risk event, the corrected evaluation result and the preset accuracy, retrain the black box analysis model in the target decision model to obtain an updated black box analysis model, wherein the black box analysis model is based on a fault prediction model of a neural network, which is used to analyze the data to be analyzed to obtain a black box analysis result, and the accuracy of the black box analysis result obtained by the updated black box analysis model for analyzing the data to be analyzed is not less than the preset accuracy; based on each historical risk event
  • the original equipment data corresponding to the risk event, the corrected evaluation results and the specified accuracy are used to adjust the judgment threshold of the fault identification rule in the white-box analysis model to obtain an updated white-box analysis model, wherein the white-box analysis model is a model based
  • the algorithm decision module is further configured to: analyze the current data to be analyzed through a white box analysis model to obtain a first analysis result, wherein the white box analysis model is a model based on the fault identification rule of the target device, and the target decision model includes the white box analysis model; analyze the current data to be analyzed through a black box analysis model to obtain a second analysis result, wherein the black box analysis model is a fault prediction model based on a neural network, and the target decision model includes the black box analysis model; based on the first analysis result and the second analysis result, as well as the target decision rule, obtain a target monitoring result corresponding to the current data to be analyzed, wherein the target decision model includes a target decision rule, and the target decision rule is used to determine a monitoring result based on the analysis result of the data by each analysis model in the target decision model, and the monitoring result is used to indicate the failure risk of the target device.
  • the algorithm expert terminal 26 is deployed in the cloud.
  • the algorithm decision end 24 is deployed on the edge side.
  • This embodiment realizes a decision-making system that is cloud-edge-integrated and uses an algorithm model as a driving force and can realize full-link closed-loop upgrade iteration, and is suitable for many industrial intelligent production application scenarios; a hierarchical and graded directive data acquisition method switching is realized between the algorithm decision end and the acquisition terminal, and the frequency of data acquisition is increased when the equipment is at high risk to ensure the validity of the data, and the frequency of data acquisition is reduced when the equipment is at low risk, which saves memory computing power and improves system stability; the closed-loop iterative upgrade and deployment application of the algorithm model based on the accuracy of the prediction results is realized between the algorithm expert end and the algorithm decision end.
  • the algorithm is self-feedback, self-adaptive, and self-learning.
  • the combination of cloud and edge ensures that the fault data is transmitted to the algorithm expert end in the cloud to the maximum extent, providing data for the algorithm correction and optimization of the target decision model. After optimization and upgrading, it is transmitted back to the algorithm decision end to form a closed-loop algorithm upgrade, and different equipment and different working conditions can be upgraded to the most suitable algorithm through this closed-loop method.
  • an electronic device 800 is also provided, which is used to implement the above-mentioned device monitoring method.
  • the electronic device 800 may include: a processor 1501, a communication interface 1502, a memory 1503 and a communication bus 1504, wherein the processor 1501, the communication interface 1502, and the memory 1503 communicate with each other through the communication bus 1504.
  • Memory 1503 configured to store computer programs
  • the processor 1501 is configured to implement the steps of the method embodiment of the present disclosure when executing the program stored in the memory 1503.
  • the bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus.
  • PCI Peripheral Component Interconnect
  • EISA Extended Industry Standard Architecture
  • the bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.
  • the communication interface is used for communication between the above electronic device and other devices.
  • the memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage.
  • RAM random access memory
  • NVM non-volatile memory
  • the memory may also be at least one storage device located away from the aforementioned processor.
  • processors can be general-purpose processors, including central processing units (CPU), network processors (NP), etc.; they can also be digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
  • CPU central processing units
  • NP network processors
  • DSP digital signal processors
  • ASIC application specific integrated circuits
  • FPGA field programmable gate arrays
  • the structure shown in FIG8 is for illustration only and does not limit the structure of the electronic device.
  • the electronic device may include more or fewer components (such as a network interface, a display device, etc.) than those shown in FIG8 , or may have a different configuration than that shown in FIG8 .
  • the embodiment of the present disclosure also provides a computer-readable storage medium, wherein the storage medium includes a stored program, wherein when the program is run, the method steps of the embodiment of the present disclosure are executed.
  • the storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.
  • ROM read-only memory
  • RAM random access memory
  • mobile hard disk a magnetic disk or an optical disk, and other media that can store program codes.
  • the present disclosure provides a device monitoring solution.
  • the acquired device data is analyzed through a target decision model to obtain real-time monitoring results of the target device, and the device data acquisition method is dynamically determined based on the real-time monitoring results of the target device, thereby achieving the purpose of automatically optimizing the data acquisition method according to the current actual situation of the target device, thereby solving the technical problem that the device monitoring system in the related technology cannot automatically optimize the data acquisition method, and achieving the technical effect of improving the monitoring capability of the device monitoring system.
  • the examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.
  • the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium.
  • the technical solution of the present disclosure, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the disclosed client can be implemented in other ways.
  • the device embodiments described above are only schematic.
  • the division of the units is only a logical function division.
  • multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.
  • Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

Abstract

公开了设备监测方法和系统、电子设备、存储介质，其中，该方法包括：按照当前数据获取方式，获取目标设备的当前待分析数据，其中，当前待分析数据是对目标设备采集得到的目标数据类型的数据；利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；按照更新后数据获取方式获取最新待分析数据，并继续对目标设备进行监测。

Description

设备监测方法和系统、电子设备、存储介质
相关申请的引用
本公开要求于2022年9月30日向中华人民共和国国家知识产权局提交的申请号为202211215947.5、发明名称为“设备监测方法和系统、电子设备、存储介质”的发明专利的优先权，并通过引用的方式将其全部内容并入本公开。
领域
本公开大体上涉及云计算技术领域，更具体地涉及设备监测方法和系统、电子设备、存储介质。
背景
随着新一轮科技革命和产业变革快速发展，工业经济由数字化向网络化、智能化深度拓展，互联网创新发展与新工业革命形成历史性交汇，催生了工业互联网。工业领域的大型关键设备价值高，结构复杂，维护成本高，一旦发生故障，损失惨重。目前，相关技术中通过在线监测系统，可进行设备故障识别、预测性维护。在线监测系统通过按照一定频率获取边缘采集设备采集得到的设备数据，根据设备数据进行设备健康状态评估与故障识别。
概述
第一方面，本公开提供了设备监测方法，其包括：按照当前数据获取方式，获取目标设备的当前待分析数据，其中，当前待分析数据是对目标设备采集得到的目标数据类型的数据；利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；按照更新后数据获取方式获取最新待分析数据，并继续对目标设备进行监测。
第二方面，本公开还提供了设备监测系统，其包括采集终端和算法决策端：采集终端，配置为对目标设备采集得到目标数据类型的待分析数据；算法决策端，配置为按照当前数据获取方式，从采集终端获取当前待分析数据，其中，当前待分析数据是采集终端对目标设备采集得到的目标数据类型的数据；利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；按照更新后数据获取方式从采集终端获取最新待分析数据，并继续对目标设备进行监测。
第三方面，本公开还提供了存储介质，该存储介质包括存储的程序，程序运行时执行本公开的方法。
第四方面，本公开还提供了电子设备，其包括：处理器、通信接口、存储器和通信总线，其中，所述处理器、通信接口和存储器通过通信总线完成相互间的通信；所述存储器，用于存放计算机程序；所述处理器，用于执行所述计算机程序时，实现本公开所述的方法。
第五方面，本公开提供了计算机可读存储介质，所述存储介质包括存储的程序，其中，所述程序运行时执行本公开所述的方法。
在某些实施方案中，本公开采用按照当前数据获取方式，获取目标设备的当前待分析数据，其中，当前待分析数据是对目标设备采集得到的目标数据类型的数据；利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；按照更新后数据获取方式获取最新待分析数据，并继续对目标设备进行监测的方式，通过目标决策模型对获取的设备数据进行分析，得到目标设备的实时监测结果，并根据对目标设备的实时监测结果，动态地确定设备数据的获取方式，达到了根据目标设备的当前实际情况自动优化数据获取方式的目的，进而解决了相关技术中的设备监测系统无法自动优化数据获取方式的技术问题，实现了提升设备监测系统的监测能力的技术效果。
附图简要说明
此处所说明的附图用来提供对本公开的进一步理解，构成本公开的一部分，本公开的示意性实施例及其说明用于解释本公开，并不构成对本公开的不当限定。在附图中：
图1为本公开一实施例提供的设备监测方法的硬件环境的示意图；
图2为公开一实施例提供的可选的设备监测方法的流程图；
图3为本公开一实施例提供的可选的设备监测系统整体功能及通讯框架示意图；
图4为本公开一实施例提供的可选的设备监测系统的采集终端与算法决策系统通讯框架示意图；
图5为本公开一实施例提供的可选的设备监测系统的算法决策系统与算法专家系统通讯框架示意图；
图6为本公开一实施例提供的可选的设备监测系统的示意图；
图7为本公开一实施例提供的另一可选的设备监测系统的示意图；以及
图8是本公开一实施例提供的电子设备的结构示意图。
详述
为了使本技术领域的人员更好地理解本公开方案，下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分的实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本公开保护的范围。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本公开的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
根据本公开实施例的一方面，本公开提供了设备监测的方法实施例。
在某些实施方案中，本公开的设备监测方法可以应用于如图1所示的由终端101和服务器103所构成的硬件环境中。如图1所示，服务器103通过网络与终端101进行连接，可用于为终端或终端上安装的客户端提供设备监测服务，可在服务器上或独立于服务器设置数据库105，用于为服务器103提供数据存储服务，上述 网络包括但不限于：广域网、城域网或局域网，终端101并不限定于PC、手机、平板电脑等。本公开实施例的设备监测方法可以由服务器103来执行，也可以由终端101来执行，还可以是由服务器103和终端101共同执行。其中，终端101执行本公开实施例的设备监测方法也可以是由安装在其上的客户端来执行。后续以在服务器上执行本公开的设备监测方法为例进行说明。
图2是根据本公开实施例的一可选的设备监测方法的流程图，如图2所示，该方法可以包括以下步骤：
步骤S202，按照当前数据获取方式，获取目标设备的当前待分析数据，其中，当前待分析数据是对目标设备采集得到的目标数据类型的数据；
步骤S204，利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；
步骤S206，确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；以及
步骤S208，按照更新后数据获取方式获取最新待分析数据，并继续对目标设备进行监测。
通过上述步骤S202至步骤S208，通过按照当前数据获取方式，获取目标设备的当前待分析数据，其中，当前待分析数据是对目标设备采集得到的目标数据类型的数据；利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；按照更新后数据获取方式获取最新待分析数据，并继续对目标设备进行监测，可以解决相关技术中的设备监测系统无法自动优化数据获取方式的技术问题，进而达到提升设备监测系统的监测能力的技术效果。
本公开的设备监测方法可适用于多种智能化决策生产场景，包括但不仅限于：设备故障识别预测、设备产能产量预测及设备生产排班、现场配件生命周期及库存量预测等。
在步骤S202提供的技术方案中，服务器按照当前数据获取方式，获取目标设备的当前待分析数据，其中，当前待分析数据是对目标设备采集得到的目标数据类型的数据；
待分析数据是对目标设备采集得到的目标数据类型的数据，可以通过对待分析数据进行分析，得到关于目标设备的设备情况的监测结果。当前待分析数据是指在当前时刻之前未被获取的、用于分析目标设备的设备情况的数据。
目标数据类型是与设备监测的目的相对应的数据类型，例如，设备监测的目的是设备故障识别预测，则目标数据类型为用于设备故障识别的数据类型，如，设备温度数据、设备压力数据、设备震动数据、设备排量数据等。
获取目标设备的当前待分析数据，可以是从采集终端获取，采集终端是指具备硬件传感器(包括但不仅限于：压力、温度、振动、排量、功率等)，以及数据采集软件(即按照一定采集频率进行数字信号采集上传的软件系统)的终端。
采集终端用于对目标设备直接采集目标数据类型的数据，采集终端具有的各类传感器安装在目标设备重要易损件附近监测点，通过电源供电，实现各测点各类传感器数据实时采集，并在其软件端通过网络将数据向外发送，可以将数据发送至服务器、其他终端、云端等。采集终端在功能上可以实现通过硬件传感器进行数据采集和通过采集软件进行数字信号解析上传，具体软硬件是否为集成或者单独部署，形式不限。
在某些实施方案中，服务器按照当前数据获取方式，获取目标设备的当前待分析数据，可以是按照当前数据获取方式，从采集终端获取目标设备的当前待分析数据。
在某些实施方案中，数据获取方式可以包括第一获取方式、第二获取式。第 一获取方式为按照第一周期获取数据，第二获取式为按照第二周期获取数据，第二周期短于所述第一周期。
在某些实施方案中，第一获取方式可以为问答式，问答式为：向采集终端发送获取指令(可以按照时间周期长度T1，等间隔发送获取指令)，采集终端响应于获取指令，发送当前待分析数据。问答式的特点：可以降低原始数据量堆积，进行等时间间隔数据采集通讯，采集频率相对较低。
在某些实施方案中，第二获取式可以为上报式，上报式为：采集终端按照预设频率主动上报当前待分析数据。例如，采集终端可以通过mqtt、文件ftp、tcp、upd等协议将待分析数据，按照时间周期长度T2，等长度等间隔、高速、实时上传至服务器、其他终端、云端等，其中，T2小于T1。上报式的特点：数据零延迟高频率实时采集，保证准确实时捕获数据中的设备故障特征。
在某些实施方案中，在首次对目标设备的待分析数据进行获取的情况下，可以将第一获取方式作为当前获取方式，即以较低频率进行数据获取，因为在刚刚开始对目标设备进行监测时，目标设备通常还是正常状态，较低频率的数据获取就已经能够满足监测需求了。
数据获取方式的种类和数量不限，获取数据的时间周期可以根据实际需要确定，不同的数据获取方式可以具有不同的时间周期。
数据传输可通过包括但不仅限于Modbus-tcp、EtherNet/IP、7S、scokect、http等协议。
在步骤S204提供的技术方案中，服务器利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果。
目标决策模型的种类和形式不限，可以为基于预设规则的判断模型，即白盒分析模型，也可以为基于神经网络的预测模型，即黑盒分析模型，也可以为二者的结合，目标决策模型中可以包含多个模型，以及根据多个模型的分析结果决策得到监测结果的决策规则。例如，对于目标监测类型为故障风险监测的场景，目标决策模型可以为基于目标设备的故障识别规则的模型，也可以为基于神经网络的故障预测模型，也可以为二者的结合。
对于目标监测类型为故障风险监测的场景，白盒分析模型包括但不限于以下三种：A.计算各类信号规定等时间间隔n或者等时间间隔n对应角域等角度间隔原始数据对应的均方误差(有效值)、均值、绝对平方值、方差、标准差、峰值、峰峰值、最大值、最小值、波形指标、脉冲指标、裕度、峭度等，通过设置自定义阈值，生成上述指标时序波动超阈值分级报警；B.基于各测点各类传感器在不同工况下的正常状态运行数据，按照等时间间隔n或者等时间间隔n对应角域等角度间隔进行抽样，堆叠所有样本数据，获得设备各测点不同工况下的正常状态信号分布基准空间，并设置基础空间超出范围阈值或规则，对待测数据故障状态进行识别，分级报警；C.基于专家经验，对指定测点位置的温度、压力、排量等热力学指标，设置设备正常状态的指标数据范围区间或者阈值，根据正常状态数据范围或者阈值判断当前待测设备该信号数据是否超出范围或者阈值存在故障风险，分级报警，例如：柱塞泵减速箱正常工作温度最高110℃，向上最大波动范围不超过20℃，则设置柱塞泵减速箱正常状态下温度区间为[110℃,130℃]，根据实际温度是否超出该范围进行故障风险报警提示。
等时间间隔n或者等时间间隔n对应角域等角度间隔：旋转类设备中单时间周期对应单角度周期，即单次运行时间周期内，对应曲轴旋转一周，即0-360°所以该单时间周期可以对应到一个完整角度周期；往复类设备，活塞杆或者拉杆往复运动一个来回对应的单时间周期也就是拉杆连接的曲轴旋转一周的时间，即0-360°。所以单时间周期也是对应一个单角度周期。即所有设备单位时间周期都可以换算到单角度周期上作为分析或者评价的刻度尺。定转速情况下：等角度周期与等时间间隔对应数据长度不变，是固定的；变转速情况下：需要根据不同时间段转速值对应时间点，进行不同转速下，时间间隔截取，即通过转速计算单周期 时间长度，然后按照不同转速下单周期时间长度对时域数据进行截断，对不同转速下不同时间长度单周期数据按照0°～360°等角度间隔进行分割即可获得单周期时域数据和对应角度域数据。
堆叠所有样本数据：例如，每个单位时间周期或者单位角度周期内，索引对应0～99，共100个点。将n个单时间周期或者单角度周期数据0～99每一个索引对应的幅值进行求平均值、最大值、3/4分位数、上包络值等统计指标中的一个，连接0～99各索引下的统计指标即可得到信号分布基准空间。
对于目标监测类型为故障风险监测的场景，黑盒分析模型包括但不限于以下两种：A.基于自监督或者无监督神经网络的设备故障识别模型，该类模型网络结构主要以自监督或者无监督神经网络为主，包括但不仅限于AE自编码器、SAE等自编码器变体，其中输入数据进入编码器进行降维特征提取，编码器结果输入解码器进行原数据还原，并通过计算解码结果与输入数据的差异性，通过设置输入与输出数据误差阈值进行设备当前故障风险等级预测，分级预警；B.基于有监督神经网络的设备故障预测模型，该类模型主要网络结构以有监督神经网络为主，包括但不仅限于CNN、RNN、LSTM等有监督神经网络模型及其变体，其中输入数据包括但不仅限于：各类信号等时间间隔或者等角度间隔抽样时域、频域数据或时频域对应的各类特征指标数据(包括但不仅限于均方误差(有效值)、均值、绝对平方值、方差、标准差、峰值、峰峰值、最大值、最小值、波形指标、脉冲指标、裕度、峭度等)、时频图或者时频矩阵等；输出为设备当前风险等级标签，标签类型对应设备风险预警等级即：设备状态正常、设备故障中风险、设备故障高风险。
目标决策模型也可以为白盒分析模型和黑盒分析模型的结合：通过白盒分析模型对当前待分析数据进行分析，得到第一分析结果，其中，白盒分析模型是基于目标设备的故障识别规则的模型，目标决策模型包括白盒分析模型；通过黑盒分析模型对当前待分析数据进行分析，得到第二分析结果，其中，黑盒分析模型是基于神经网络的故障预测模型，目标决策模型包括黑盒分析模型；基于第一分析结果和第二分析结果，以及目标决策规则，得到与当前待分析数据对应的目标监测结果，其中，目标决策模型包括目标决策规则，目标决策规则用于根据目标决策模型中每个分析模型对数据的分析结果确定出监测结果，监测结果用于指示目标设备的故障风险。
目标监测类型包括但不限于以下几种：设备故障风险监测、设备产能产量预测、设备生产排班、现场配件生命周期及库存量预测。
监测结果的形式不限，可以为标签结果、数值结果等。例如，对于设备故障风险监测的场景，监测结果可以为指示目标设备的故障风险程度的标签，如，“设备正常”、“设备低风险”、“设备中风险”、“设备高风险”；监测结果可以为指示目标设备的故障风险程度的数值，数值越高表示故障风险越高。
在步骤S206提供的技术方案中，服务器确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式。
对于设备监测来说，针对目标设备的不同情况，可以采取不同的数据获取方式，以达到更好的监测效果。例如，对于设备故障风险监测的场景，若目标设备当前的故障风险等级为中高风险，则需要以更高的数据获取频率获取数据，准确实时捕获数据中的设备故障特征，以得到更加精准监测结果；若目标设备当前的故障风险等级为零风险或低风险，则无需以较高的数据获取频率获取数据，如果仍以较高的数据获取频率获取数据，可能会导致数据堆积、算力浪费等问题。
服务器可以根据目标监测结果得知目标设备的当前情况，以确定出与当前情况相符的数据获取方式，并按照更新后数据获取方式进行数据获取。例如，目标监测结果指示目标设备的故障风险为高风险，则需要将当前获取方式由周期较长的第一获取方式更新为周期较短的第二获取方式。
在步骤S208提供的技术方案中，服务器按照更新后数据获取方式获取最新待 分析数据，并继续对目标设备进行监测。
对目标设备进行监测是指获取目标设备的数据并分析得到监测结果，以及根据监测结果确定数据获取方式的过程，对设备的监测是长期的监测，也是实时的监测，即在监测过程中不断获取最新的待分析数据，不断分析得到待分析数据对应的监测结果，也不断地根据最新的监测结果确定出适应于目标设备的最新情况的数据获取方式。
在更新数据获取方式后，即按照更新后数据获取方式获取最新待分析数据，并利用目标决策模型对最新待分析数据进行分析得到最新监测结果，并根据最新监测结果确定出最新的数据获取方式。
在某些实施方案中，目标监测类型为故障风险监测，步骤S206中，确定出与目标监测结果对应的目标数据获取方式，还包括如下所述的步骤：
步骤S31，在当前数据获取方式为按照第一周期获取数据的第一获取方式，且目标监测结果指示目标设备的故障风险小于或等于风险下限的情况下，确定出目标数据获取方式为第一获取方式；以及
步骤S32，在当前数据获取方式为第一获取方式，且目标监测结果指示目标设备的故障风险大于风险下限的情况下，确定出目标数据获取方式为按照第二周期获取数据的第二获取方式，其中，第二周期短于第一周期。
第一获取方式可以为问答式，问答式为：向采集终端发送获取指令(可以按照时间周期长度T1，等间隔发送获取指令)，采集终端响应于获取指令，发送当前待分析数据。问答式的特点：可以降低原始数据量堆积，进行等时间间隔数据采集通讯，采集频率相对较低。
第二获取式可以为上报式，上报式为：采集终端按照预设频率主动上报当前待分析数据。例如，采集终端可以通过mqtt、文件ftp、tcp、upd等协议将待分析数据，按照时间周期长度T2，等长度等间隔、高速、实时上传至服务器、其他终端、云端等，其中，T2小于T1。上报式的特点：数据零延迟高频率实时采集，保证准确实时捕获数据中的设备故障特征。
在首次对目标设备的待分析数据进行获取的情况下，可以将第一获取方式作为当前获取方式，即以较低频率进行数据获取，因为在刚刚开始对目标设备进行监测时，目标设备通常还是正常状态，较低频率的数据获取就已经能够满足监测需求了。
在当前数据获取方式为第一获取方式，且目标监测结果指示目标设备的故障风险大于风险下限的情况下，则说明目标设备目前未出现异常，按照原来的较低频率获取数据即可，无需变更数据获取方式，即确定出目标数据获取方式为按照第一周期获取数据的第一获取方式。
在当前数据获取方式为第一获取方式，目标监测结果指示目标设备的故障风险大于风险下限的情况下，则说明目标设备目前开始出现异常，可能有故障，需要高频获取数据，以保证准确实时捕获数据中的设备故障特征，即确定出目标数据获取方式为按照第二周期获取数据的第二获取方式。
默认采用以较低频率获取数据的数据获取方式，当目标监测结果首次指示目标设备的故障风险大于风险下限，则改用以较高频率获取数据的数据获取方式。
在某些实施方案中，在步骤S204中，利用目标决策模型对当前待分析数据进行分析，得到目标监测结果之后，该方法还包括如下所述的步骤：
步骤S321，在当前数据获取方式为第一获取方式，且目标监测结果指示目标设备的故障风险大于风险下限的情况下，生成指定风险事件，其中，指定风险事件用于指示目标设备的故障风险由小于或等于风险下限转变为大于风险下限。
在当前数据获取方式为第一获取方式，目标监测结果指示目标设备的故障风险大于风险下限的情况下，则说明目标设备目前开始出现异常，可能有故障，需要生成风险事件，并在数据库表中记录该风险事件的相关数据，以便后续进一步判断风险、采取风险控制策略。
在某些实施方案中，步骤206，确定出与目标监测结果对应的目标数据获取方 式，还包括如下所述的步骤：
步骤S322，确定出与目标监测结果对应的目标数据获取方式为第二获取方式，通过目标数据获取方式将当前数据获取方式由第一获取方式更新为第二获取方式。
在当前数据获取方式为第一获取方式，目标监测结果指示目标设备的故障风险大于风险下限的情况下，步骤206即按照步骤S322的方式执行。
在某些实施方案中，步骤208，按照更新后数据获取方式获取最新待分析数据，并继续对目标设备进行监测，还包括如下所述的步骤：
步骤S323，按照第二获取方式获取最新待分析数据并继续对目标设备进行监测。
在当前数据获取方式为第一获取方式，目标监测结果指示目标设备的故障风险大于风险下限的情况下，步骤208即按照步骤S323的方式执行。
在某些实施方案中，步骤323，按照第二获取方式获取最新待分析数据并继续对目标设备进行监测，还包括执行如下所述的循环步骤：
步骤S3231，按照第二获取方式，获取目标设备的最新待分析数据，其中，最新待分析数据是对目标设备采集得到的目标数据类型的数据；以及
步骤S3232，利用目标决策模型对最新待分析数据进行分析，得到最新监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果。
在将数据获取方式更新为第二获取方式之后，可以不再在设备监测过程中进行确定数据获取方式的步骤，因为此时目标设备已经出现了异常，即使后续的监测结果指示目标设备正常，也不能排除目标设备的故障风险，因此，需要继续按照第二获取方式进行数据获取。
如果经确认，该目标设备实际上是正常的，那么也可以将数据获取方式由第二获取方式变更为第一获取方式，继续循环执行完整的设备监测步骤，在监测过程中不断获取最新的待分析数据，不断分析得到待分析数据对应的监测结果，也不断地根据最新的监测结果确定出适应于目标设备的最新情况的数据获取方式，按照最新变更后数据获取方式进行后续监测过程中的数据获取。
在某些实施方案中，在步骤208，按照更新后数据获取方式获取最新待分析数据之后，该方法还包括如下所述的步骤：
步骤S41，获取与指定时间段内获取的每个最新待分析数据对应的最新监测结果，其中，与每个最新待分析数据对应的最新监测结果是利用目标决策模型对最新待分析数据进行分析得到的；以及
步骤S42，按照目标监测结果以及最新监测结果，确定出与指定风险事件对应的风险控制策略。
针对目标设备首次出现监测结果指示故障风险大于风险下限的情况，采集终端以上报式为更新后数据获取方式，后续的监测过程中，采集终端以周期长度T等时间间隔上报最新待分析数据至服务器，服务器对每次上报的最新待分析数据进行分析得到与每次上报对应的最新监测结果。
服务器获取指定时间段t内，每一次获取的最新待分析数据对应的最新监测结果，得到t/T个时序的连续监测结果，T为数据获取周期长度。
按照目标监测结果和指定时间段内所有最新监测结果(t/T个时序的连续监测结果)，确定出与指定风险事件对应的风险控制策略。监测结果的分级规则和风险控制策略规则确定方式如下所示：
Figure PCTCN2022133182-appb-000001
Figure PCTCN2022133182-appb-000002
在某些实施方案中，在步骤S41，在获取与指定时间段内获取的每个最新待分析数据对应的最新监测结果之后，方法还包括：
步骤S43，将目标监测结果以及最新监测结果，作为指定风险事件的算法评价结果，并将对应于目标监测结果以及最新监测结果的所有目标待分析数据，作为指定风险事件的原始设备数据，存储至历史风险事件库中，其中，目标待分析数据包括当前待分析数据和最新待分析数据；以及
步骤S44，获取目标对象对指定风险事件进行评价得到的指定评价结果，并将指定评价结果作为指定风险事件的实际评价结果，存储至历史风险事件库，其中，实际评价结果用于指示所对应的风险事件实际认定的故障风险。
服务器可以从专家评价接口获取对风险事件的实际评价结果，实际评价结果是专业人员经过对目标设备的实际检修和专业分析后的得到的人工评价，评价模式不限，可以为选择标签模式，也可以输入评分模式，例如，评价模式为三值单选，即“确实高风险故障”、“确实中风险故障”或“确实设备正常”。
服务器会将每个风险事件的原始设备数据、算法评价结果、实际评价结果进行保存到后台对应数据库表生成历史风险事件记录表，以便后续进行目标决策模型的准确度评估和更新。
在某些实施方案中，该方法还包括如下所述的步骤：
步骤S51，从历史风险事件库中获取多个历史事件数据，其中，每个历史事件数据唯一对应于一个历史风险事件，每个历史风险事件对应的历史事件数据包括与每个历史风险事件对应的原始设备数据、算法评价结果以及实际评价结果；
步骤S52，按照每个历史事件数据中的算法评价结果和实际评价结果，确定出目标决策模型的当前准确度；
步骤S53，在目标决策模型的当前准确度低于预设阈值的情况下，根据多个历史事件数据对目标决策模型进行更新，得到更新后的目标决策模型。
目标决策模型的当前准确度＝(算法评价结果与实际评价结果相符的历史风险事件数量)/历史风险事件总数量。
可以通过设置目标决策模型准确率阈值α，判断当前模型预测准确率是否满足要求。如果目标决策模型预测准确率小于α，可以对目标决策模型进行调整、更新。
在某些实施方案中，步骤S53，根据多个历史事件数据对目标决策模型进行更新，得到更新后的目标决策模型，还包括如下所述的步骤：
步骤S531，对于每个历史事件数据对应的历史风险事件，按照历史事件数据 中的实际评价结果，对的历史事件数据中的算法评价结果进行修正，得到历史风险事件的修正评价结果；
步骤S532，基于每个历史风险事件对应的原始设备数据、修正评价结果以及预设准确率，对目标决策模型中的黑盒分析模型进行重训练，得到更新后黑盒分析模型，其中，黑盒分析模型基于神经网络的故障预测模型，用于对待分析数据进行分析得到黑盒分析结果，更新后黑盒分析模型对待分析数据进行分析得到的黑盒分析结果的准确率不低于预设准确率；
步骤S533，基于每个历史风险事件对应的原始设备数据、修正评价结果以及指定准确率，调整白盒分析模型中故障识别规则的判断阈值，得到更新后白盒分析模型，其中，白盒分析模型是目标决策模型中基于故障识别规则的模型，用于根据故障识别规则对待分析数据进行分析得到白盒分析结果，更新后白盒分析模型对待分析数据进行分析得到的白盒分析结果的准确率不低于指定准确率；以及
步骤S534，基于目标决策模型中的目标决策规则、更新后黑盒分析模型以及更新后白盒分析模型，得到更新后的目标决策模型，其中，目标决策规则用于根据目标决策模型中每个分析模型对数据的分析结果确定出监测结果。
白盒分析模型的更新可以借助领域专家进行人工经验分析，确定白盒分析模型误判根因，并根据专家经验对白盒分析模型阈值进行适当调整修正，使得修正以后的白盒分析模型预测准确率提升至指定准确率。
在某些实施方案中，目标监测类型为故障风险监测，步骤S204，利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，还包括如下所述的步骤：
步骤S21，通过白盒分析模型对当前待分析数据进行分析，得到第一分析结果，其中，白盒分析模型是基于目标设备的故障识别规则的模型，目标决策模型包括白盒分析模型；
步骤S22，通过黑盒分析模型对当前待分析数据进行分析，得到第二分析结果，其中，黑盒分析模型是基于神经网络的故障预测模型，目标决策模型包括黑盒分析模型；以及
步骤S23，基于第一分析结果和第二分析结果，以及目标决策规则，得到与当前待分析数据对应的目标监测结果，其中，目标决策模型包括目标决策规则，目标决策规则用于根据目标决策模型中每个分析模型对数据的分析结果确定出监测结果，监测结果用于指示目标设备的故障风险。
在某些实施方案中，根据白盒分析模型和黑盒分析模型进行双轨混合故障识别，目标决策规则如下所示：
Figure PCTCN2022133182-appb-000003
Figure PCTCN2022133182-appb-000004
通过双轨混合分析，可以提高监测结果的准确度，更加灵敏地识别出目标设备的故障风险，提升了设备监测系统的监测能力。
在某些实施方案中，下文结合实施方式示意性的描述本公开的技术方案：
近年来，新一轮科技革命和产业变革快速发展，工业经济由数字化向网络化、智能化深度拓展，互联网创新发展与新工业革命形成历史性交汇，催生了工业互联网。工业领域的大型关键设备价值高，结构复杂，维护成本高，一旦发生故障，损失惨重。通过算法决策系统架构构建的在线监测系统，可进行设备故障识别、预测性维护，不仅降低了设备运维成本，还提高了生产资源的动态配置效率。算法决策系统设置一系列传感器以实时采集设备数据，根据设备运行的历史数据，建立设备的决策算法模型，再利用此算法模型实时计算设备数据，从而进行设备健康状态评估与故障识别。
在线监测系统一般包括三部分模块：边缘采集设备、边缘算法决策系统、云端算法专家系统，在相关技术中三部分模块是独立执行没有关联的。其中边缘采集设备的采集频率、传输方式一经设定完毕后在运行过程中是不可变的，如果采集频率设定过高则会造成内存算力的浪费，而设定过低则会不利于数据分析故障诊断；边缘算法决策系统则集成了大量决策算法进行实时决策，而算法集成完毕后几乎不再升级或者升级困难；云端算法专家系统中的算法一般是分析大量数据后提炼而成，但算法提炼完毕后就不再优化升级或者因为缺少故障数据无法升级，从而导致算法只适应某一设备无法自适应升级。
本实施例提供的云边结合的算法闭环决策系统架构方法，以及一种分级决策识别设备的风险等级、且决策算法自动闭环升级优化的系统，可联动三部分模块，云边协同工作，动态采集数据，自动升级决策算法。
本实施例可适用于多种智能化决策生产场景，包括但不仅限于：设备故障识别预测、设备产能产量预测及设备生产排班、现场配件生命周期及库存量预测等。
本实施例主要以设备故障识别预测场景为例进行方法阐述。以设备故障预测场景为例，系统包括三部分模块：边缘采集终端、边缘算法决策系统、云端算法专家系统。图3是本公开一实施例的可选的设备监测系统整体功能及通讯框架示意图。设备正常状态下，边缘端系统通过问答方式与采集终端进行通讯，获取并存储原始传感器信号数据；边缘端系统部署白盒类规则模型和黑盒类深度神经网络模型，进行双轨混合故障识别分级预警；边缘端系统根据故障识别预警等级进行与终端数据通讯方式切换；边缘端系统根据实时且持续诊断结果，发送分级控制指令；边缘端pc系统及app设置故障识别模型预测准确率人工评判入口，对风险事件模型预测准确率进行分级评价；边缘端系统将历史故障预警事件、预警事件涉及传感器原始数据及该事件预警准确率人工评判结果，通过网络回传到云端算法专家系统，算法专家系统根据算法历史准确率及预警事件原始数据，重新训练升级模型；云端算法专家系统完成模型升级迭代后，将算法模型远程通过网络传输给边缘端系统，进行模型替换迭代，完成云边结合的全闭环算法决策功能实现。
本实施例所述设备数据采集终端，下称采集终端，指一类具备硬件传感器(包括但不仅限于：压力、温度、振动、排量、功率等)，以及数据采集软件(即按照一定采集频率进行数字信号采集上传的软件系统)的终端系统。各类传感器安装在目标设备重要易损件附近监测点，通过电源供电，实现各测点各类传感器数据实时采集，并在其软件端通过网络将数据向外发送。
本实施例所述边缘端算法决策系统，下称算法决策系统，指以白盒和黑盒两 类算法模型服务为主体进行设备当前状态故障风险识别的软件系统。该系统输入为从设备数据采集终端获取的各类采集信号原始数据，输出为当前设备采集信号对应测点设备部件的故障风险等级。
在默认状态下，采集终端和算法决策系统之间通过问答方式进行数据通讯，由算法决策系统等时间间隔主动去获取采集终端最新数据，数据传输可通过但不仅限于：Modbus-tcp、EtherNet/IP、7S、scokect、http等协议。
上述算法决策系统，主要包括两部分模型服务即白盒、黑盒算法模型服务程序。
其中，白盒算法模型服务指的是一类故障识别规则算法模型，规则算法模型是指，该类规则算法模型包括但不限于以下三种：
1.计算各类信号规定等时间间隔n或者等时间间隔n对应角域等角度间隔原始数据对应的均方误差(有效值)、均值、绝对平方值、方差、标准差、峰值、峰峰值、最大值、最小值、波形指标、脉冲指标、裕度、峭度等，通过设置自定义阈值，生成上述指标时序波动超阈值分级报警；
2.基于各测点各类传感器在不同工况下的正常状态运行数据，按照等时间间隔n或者等时间间隔n对应角域等角度间隔进行抽样，堆叠所有样本数据，获得设备各测点不同工况下的正常状态信号分布基准空间，并设置基础空间超出范围阈值或规则，对待测数据故障状态进行识别，分级报警；
3.基于专家经验，对指定测点位置的温度、压力、排量等热力学指标，设置设备正常状态的指标数据范围区间或者阈值，根据正常状态数据范围或者阈值判断当前待测设备该信号数据是否超出范围或者阈值存在故障风险，分级报警；
另外，黑盒算法模型服务指的是一类基于深度神经网络的故障识别模型，该类神经网络算法模型包括但不仅限于以下两种：
A.基于自监督或者无监督神经网络的设备故障识别模型，该类模型网络结构主要以自监督或者无监督神经网络为主，包括但不仅限于AE自编码器、SAE等自编码器变体，其中输入数据进入编码器进行降维特征提取，编码器结果输入解码器进行原数据还原，并通过计算解码结果与输入数据的差异性，通过设置输入与输出数据误差阈值进行设备当前故障风险等级预测，分级预警；
B.基于有监督神经网络的设备故障预测模型，该类模型主要网络结构以有监督神经网络为主，包括但不仅限于CNN、RNN、LSTM等有监督神经网络模型及其变体，其中输入数据包括但不仅限于：各类信号等时间间隔或者等角度间隔抽样时域、频域数据或时频域对应的各类特征指标数据(包括但不仅限于均方误差(有效值)、均值、绝对平方值、方差、标准差、峰值、峰峰值、最大值、最小值、波形指标、脉冲指标、裕度、峭度等)、时频图或者时频矩阵等；输出为设备当前风险等级标签，标签类型对应设备风险预警等级即：设备状态正常、设备故障中风险、设备故障高风险。
最后，上述白盒和黑盒模型对当前待测样本数据做出预测后，将白盒模型与黑盒模型按照如下规则生成设备当前风险等级双轨识别结果。
Figure PCTCN2022133182-appb-000005
Figure PCTCN2022133182-appb-000006
图4是根据本公开实施例的一可选的设备监测系统的采集终端与算法决策系统通讯框架示意图。算法决策系统根据白盒和黑盒进行双轨混合故障识别，输出设备当前风险等级，该结果数据会通过scokect、http、Modbus-tcp、EtherNet/IP、7S等等通讯协议即时反馈给采集终端，当风险等级为设备故障中风险或高风险时，采集终端会触发数据通讯方式转换指令，将问答式通讯方式转换为主动上报式，采集终端通过mqtt、文件ftp、tcp、upd等等协议将对应中高风险测点原始信号数据，按照时间周期长度T，等长度等间隔、高速、实时发送到算法决策系统。
针对设备首报中、高风险的测点信号数据，采集终端会通过主动上报的方式将信号时域数据发送到算法决策系统，算法决策系统会将风险事件原始数据进行数据库建表存储，并持续t时间段内对每一周期长度为T的实时数据进行双轨故障识别预测，并输出t/T个时序连续预测结果。算法系统会根据连续输出的双轨故障识别预测结果，进行分级决策控制，即不同等级的决策控制命令发送，从而起到主动决策设备运行、检修的目的。控制命令发送主要以短信、移动端APP通知、PC端弹窗、大屏监控端弹窗等形式发出，主要决策控制等级及分级规则如下所示：
Figure PCTCN2022133182-appb-000007
算法决策系统会在pc端、移动app等设置故障识别模型预测准确率专家人工评价入口，每次发生风险事件，现场工作人员需要手动确认风险通知或者控制指令是否执行。并在此次风险事件后结合实际检修和专家人工评判结果，评价算法故障识别预测准确性，评价模式为三值单选，即确实高风险故障、确实中风险故障或确实设备正常。算法决策系统会将此次风险事件及专家人工评价结果保存到后台对应数据库表生成历史风险事件记录表。
边缘端算法决策系统定期将历史风险事件记录表和历史风险事件对应的原始测点信号数据通过消息中间件(如emqx、rabbitmq、kafka)或者http接口、文件ftp等等方式传输至云端算法专家系统，下称算法专家系统。算法专家系统是一套算法升级迭代和推送的支持系统，算法专家系统后台数据库存储来自算法决策系统的历史风险事件记录表和历史风险事件对应原始测点信号数据，通过对比历史风险事件风险等级标签和人工评价选项一致性，获得算法模型准确率，即：
算法模型准确率＝(风险等级标签与专家人工评价相同的风险事件数量)/历史风险事件总数量
通过设置算法模型准确率阈值α，判断当前模型预测准确率是否满足要求。如果模型预测准确率小于α,算法专家系统可以提供模型重训练功能，过程如下：
A.对历史风险事件风险等级标签根据专家人工评价选项进行修正，即用专家人工评价选项替换历史风险等级标签，完成误预测数据的修正；
B.基于历史风险事件原始测点信号数据以及修正以后的历史风险等级标签，对黑盒模型进行重训练，并根据数据情况设置模型预测精度提升率阈值β，使得重训练的黑盒模型预测准确度提升率大于β,输出完成重训练的黑盒模型；
C.基于历史风险事件进行风险等级标签的源是测点信号数据，借助领域专家进行人工经验分析，确定白盒模型误判根因，并根据专家经验对白盒模型阈值进行适当调整修正，使得修正以后的白盒模型预测准确度提升率大于β,输出完成修正的白盒模型对应阈值；
D.算法专家系统通过远程网络传输将重训练黑盒模型和完成修正的百合白盒模型阈值上传到边缘端算法决策系统，并完成模型更新部署。
图5是根据本公开实施例的一可选的设备监测系统的算法决策系统与算法专家系统通讯框架示意图。基于上述方法过程即可实现云边结合的设备故障识别算法模型决策及模型闭环修正全过程，同理该方法也可用于其他生产场景，包括但不仅限于：设备故障识别预测、生产设备产能产量预测、现场配件生命周期计算及库存量预测等。
本实施例的优点如下：
1、实现了一种云边结合的以算法模型作为驱动力并能实现全链路闭环升级迭代的决策系统，且适用于众多工业智能化生产应用场景。
2、边缘算法决策系统与信号采集终端之间实现分层分级指令式通讯切换的数据传输方法，设备处于高风险时提高采集传输速率，保证数据的有效性，设备处于低风险时降低速率节省了内存算力，提高了系统稳定性。
3、云端算法专家系统与边缘算法决策系统之间实现基于预测结果准确率的算法模型闭环迭代升级和部署应用。算法自反馈、自适应、自学习，云边结合保证了故障数据最大限度的传输至云端，为算法修正优化提供了数据，优化升级后算法再回转至边缘端，形成闭环算法升级，且不同设备、不同工况可通过此闭环方式升级为最适合的算法。
4、边缘端算法决策系统实现规则类机理白盒算法模型和基于神经网络黑盒算法模型的双轨混合决策分级控制。
需要说明的是，对于前述的各方法实施例，为了简单描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本公开并不受所描述的动作顺序的限制，因为依据本公开，某些步骤可以采用其他顺序或者同时进行。其次， 本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定是本公开所必须的。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到根据上述实施例的方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本公开各个实施例所述的方法。
根据本公开实施例的另一个方面，还提供了用于实施本公开设备监测方法的设备监测系统600。图6是根据本公开实施例的一可选的设备监测系统600的示意图，如图6所示，该系统600可以包括：
采集终端22，配置为对目标设备采集得到目标数据类型的待分析数据；
算法决策端24，配置为按照当前数据获取方式，从采集终端获取当前待分析数据，其中，当前待分析数据是采集终端对目标设备采集得到的目标数据类型的数据；利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式；按照更新后数据获取方式从采集终端获取最新待分析数据，并继续对目标设备进行监测。
需要说明的是，该实施例中的算法决策端24可以用于执行本公开实施例中的步骤S202、S204、S206、S208，该实施例中的采集终端22可以用于采集本公开实施例中的步骤S202、步骤S208中的待分析数据。
需要说明的是，上述采集终端22和算法决策端24作为系统的一部分可以运行在如图1所示的硬件环境中，可以通过软件实现，也可以通过硬件实现。
上述采集终端22和算法决策端24，通过目标决策模型对获取的设备数据进行分析，得到目标设备的实时监测结果，并根据对目标设备的实时监测结果，动态地确定设备数据的获取方式，达到了根据目标设备的当前实际情况自动优化数据获取方式的目的，进而解决了相关技术中的设备监测系统无法自动优化数据获取方式的技术问题，实现了提升设备监测系统的监测能力的技术效果。
在某些实施方案中，算法决策端24可以包括：数据获取模块，配置为按照当前数据获取方式，从采集终端获取当前待分析数据，其中，当前待分析数据是采集终端对目标设备采集得到的目标数据类型的数据；算法决策模块，配置为利用目标决策模型对当前待分析数据进行分析，得到目标监测结果，其中，目标决策模型用于对目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与目标监测结果对应的目标数据获取方式，获取方式更新模块，配置为通过目标数据获取方式对当前数据获取方式进行更新，得到更新后数据获取方式。
需要说明的是，该实施例中的数据获取模块可以用于执行本公开实施例中的步骤S202，该实施例中的算法决策模块可以用于执行本公开实施例中的步骤S204，该实施例中的获取方式更新模块可以用于执行本公开实施例中的步骤S206，该实施例中的数据获取模块、算法决策模块、获取方式更新模块可以用于共同执行本公开实施例中的步骤S208。
此处需要说明的是，上述模块与对应的步骤所实现的示例和应用场景相同，但不限于上述实施例所公开的内容。需要说明的是，上述模块作为系统的一部分可以运行在如图1所示的硬件环境中，可以通过软件实现，也可以通过硬件实现。
图7是根据本公开实施例的另一可选的设备监测系统700的示意图，如图7所示，该系统700可以包括采集终端22、算法决策端24和算法专家端26。
在某些实施方案中，在目标监测类型为故障风险监测的情况下，获取方式更新模块，还配置为：在当前数据获取方式为按照第一周期获取数据的第一获取方式，且目标监测结果指示目标设备的故障风险小于或等于风险下限的情况下，确 定出目标数据获取方式为第一获取方式；在当前数据获取方式为第一获取方式，在目标监测结果指示目标设备的故障风险大于风险下限的情况下，确定出目标数据获取方式为按照第二周期获取数据的第二获取方式，其中，第二周期短于第一周期。
在某些实施方案中，在利用目标决策模型对当前待分析数据进行分析，得到目标监测结果之后，算法决策端24可以包括事件数据模块，配置为：在当前数据获取方式为第一获取方式，且目标监测结果指示目标设备的故障风险大于风险下限的情况下，生成指定风险事件，其中，指定风险事件用于指示目标设备的故障风险由小于或等于风险下限转变为大于风险下限。
在某些实施方案中，事件数据模块，还配置为：在当前获取方式为第一获取方式，更新后数据获取方式为第二获取方式的情况下，在按照更新后数据获取方式获取最新待分析数据之后，获取与指定时间段内获取的每个最新待分析数据对应的最新监测结果，其中，与每个最新待分析数据对应的最新监测结果是利用目标决策模型对最新待分析数据进行分析得到的。
在某些实施方案中，算法决策端24还可以包括风险策略确定模块，配置为：按照目标监测结果以及最新监测结果，确定出与指定风险事件对应的风险控制策略。
在某些实施方案中，事件数据模块，还配置为：将目标监测结果以及最新监测结果，作为指定风险事件的算法评价结果，并将对应于目标监测结果以及最新监测结果的所有目标待分析数据，作为指定风险事件的原始设备数据，存储至历史风险事件库中，其中，目标待分析数据包括当前待分析数据和最新待分析数据；获取目标对象对指定风险事件进行评价得到的指定评价结果，并将指定评价结果作为指定风险事件的实际评价结果，存储至历史风险事件库，其中，实际评价结果用于指示所对应的风险事件实际认定的故障风险。
在某些实施方案中，该系统还可以包括算法专家端26，包括：历史数据获取模块，用于从历史风险事件库中获取多个历史事件数据，其中，每个历史事件数据唯一对应于一个历史风险事件，每个历史风险事件对应的历史事件数据包括与每个历史风险事件对应的原始设备数据、算法评价结果以及实际评价结果；模型评估模块，用于按照每个历史事件数据中的算法评价结果和实际评价结果，确定出目标决策模型的当前准确度；模型更新模块，用于在目标决策模型的当前准确度低于预设阈值的情况下，根据多个历史事件数据对目标决策模型进行更新，得到更新后的目标决策模型。
在某些实施方案中，模型更新模块，还配置为：对于每个历史事件数据对应的历史风险事件，按照历史事件数据中的实际评价结果，对的历史事件数据中的算法评价结果进行修正，得到历史风险事件的修正评价结果；基于每个历史风险事件对应的原始设备数据、修正评价结果以及预设准确率，对目标决策模型中的黑盒分析模型进行重训练，得到更新后黑盒分析模型，其中，黑盒分析模型基于神经网络的故障预测模型，用于对待分析数据进行分析得到黑盒分析结果，更新后黑盒分析模型对待分析数据进行分析得到的黑盒分析结果的准确率不低于预设准确率；基于每个历史风险事件对应的原始设备数据、修正评价结果以及指定准确率，调整白盒分析模型中故障识别规则的判断阈值，得到更新后白盒分析模型，其中，白盒分析模型是目标决策模型中基于故障识别规则的模型，用于根据故障识别规则对待分析数据进行分析得到白盒分析结果，更新后白盒分析模型对待分析数据进行分析得到的白盒分析结果的准确率不低于指定准确率；基于目标决策模型中的目标决策规则、更新后黑盒分析模型以及更新后白盒分析模型，得到更新后的目标决策模型，其中，目标决策规则用于根据目标决策模型中每个分析模型对数据的分析结果确定出监测结果。
在某些实施方案中，在目标监测类型为故障风险监测的情况下，算法决策模块，还配置为：通过白盒分析模型对当前待分析数据进行分析，得到第一分析结果，其中，白盒分析模型是基于目标设备的故障识别规则的模型，目标决策模型 包括白盒分析模型；通过黑盒分析模型对当前待分析数据进行分析，得到第二分析结果，其中，黑盒分析模型是基于神经网络的故障预测模型，目标决策模型包括黑盒分析模型；基于第一分析结果和第二分析结果，以及目标决策规则，得到与当前待分析数据对应的目标监测结果，其中，目标决策模型包括目标决策规则，目标决策规则用于根据目标决策模型中每个分析模型对数据的分析结果确定出监测结果，监测结果用于指示目标设备的故障风险。
在某些实施方案中，算法专家端26部署在云端。
在某些实施方案中，算法决策端24部署在边缘侧。
本实施例实现了云边结合的以算法模型作为驱动力并能实现全链路闭环升级迭代的决策系统，且适用于众多工业智能化生产应用场景；算法决策端与采集终端之间实现分层分级指令式数据获取方式切换，设备处于高风险时提高数据获取频率，保证数据的有效性，设备处于低风险时降低数据获取频率，节省了内存算力，提高了系统稳定性；算法专家端与算法决策端之间实现基于预测结果准确率的算法模型闭环迭代升级和部署应用。算法自反馈、自适应、自学习，云边结合保证了故障数据最大限度地传输至云端的算法专家端，为目标决策模型的算法修正优化提供了数据，优化升级后再回传至算法决策端，形成闭环算法升级，且不同设备、不同工况可通过此闭环方式升级为最适合的算法。
根据本公开实施例的另一个方面，还提供电子设备800，其用于实现上述设备监测方法，如图8所示，电子设备800可以包括：处理器1501、通信接口1502、存储器1503和通信总线1504，其中，处理器1501，通信接口1502，存储器1503通过通信总线1504完成相互间的通信。
存储器1503，配置为存放计算机程序；
处理器1501，配置为执行存储器1503上所存放的程序时，实现本公开的方法实施例的步骤。
上述电子设备提到的总线可以是外设部件互连标准(Peripheral Component Interconnect，PCI)总线或扩展工业标准结构(Extended Industry Standard Architecture，EISA)总线等。该总线可以分为地址总线、数据总线、控制总线等。为便于表示，图中仅用一条粗线表示，但并不表示仅有一根总线或一种类型的总线。
通信接口用于上述电子设备与其他设备之间的通信。
存储器可以包括随机存取存储器(Random Access Memory，RAM)，也可以包括非易失性存储器(Non-Volatile Memory，NVM)，例如至少一个磁盘存储器。在某些实施方案中，存储器还可以是至少一个位于远离前述处理器的存储装置。
上述的处理器可以是通用处理器，包括中央处理器(Central Processing Unit，CPU)、网络处理器(Network Processor，NP)等；还可以是数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。
本领域普通技术人员可以理解，图8所示的结构仅为示意，并不对上述电子设备的结构造成限定。例如，电子设备还可包括比图8中所示更多或者更少的组件(如网络接口、显示装置等)，或者具有与图8所示不同的配置。
本公开实施例还提供计算机可读存储介质，存储介质包括存储的程序，其中，程序运行时执行本公开实施例的方法步骤。
在某些实施方案中，在本实施例中，上述存储介质可以包括但不限于：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
在某些实施方案中，本公开提供了设备监测的方案。通过目标决策模型对获取的设备数据进行分析，得到目标设备的实时监测结果，并根据对目标设备的实时监测结果，动态地确定设备数据的获取方式，达到了根据目标设备的当前实际情况自动优化数据获取方式的目的，进而解决了相关技术中的设备监测系统无法 自动优化数据获取方式的技术问题，实现了提升设备监测系统的监测能力的技术效果。
在某些实施方案中，本实施例中的示例可以参考上述实施例中所描述的示例，本实施例在此不再赘述。
需要说明的是，在本文中，诸如“第一”和“第二”等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
上述本公开实施例序号仅仅为了描述，不代表实施例的优劣。
上述实施例中的集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在上述计算机可读取的存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在存储介质中，包括若干指令用以使得一台或多台计算机设备(可为个人计算机、服务器或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。
在本公开的上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本公开所提供的几个实施例中，应该理解到，所揭露的客户端，可通过其它的方式实现。其中，以上所描述的装置实施例仅仅是示意性的，例如所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，单元或模块的间接耦合或通信连接，可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本公开各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
以上所述仅是本公开的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本公开原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本公开的保护范围。

Claims (10)

  1. 设备监测方法，其包括：
    按照当前数据获取方式，获取目标设备的当前待分析数据，其中，所述当前待分析数据是对所述目标设备采集得到的目标数据类型的数据；
    利用目标决策模型对所述当前待分析数据进行分析，得到目标监测结果，其中，所述目标决策模型用于对所述目标数据类型的数据进行分析得到目标监测类型的监测结果；
    确定出与所述目标监测结果对应的目标数据获取方式，通过所述目标数据获取方式对所述当前数据获取方式进行更新，得到更新后数据获取方式；以及
    按照所述更新后数据获取方式获取最新待分析数据，并继续对所述目标设备进行监测。
  2. 如权利要求1所述的方法，其中，所述目标监测类型为故障风险监测，所述确定出与所述目标监测结果对应的目标数据获取方式，包括：
    在所述当前数据获取方式为按照第一周期获取数据的第一获取方式，且所述目标监测结果指示所述目标设备的故障风险小于或等于风险下限的情况下，确定出所述目标数据获取方式为所述第一获取方式；以及
    在所述当前数据获取方式为所述第一获取方式，且所述目标监测结果指示所述目标设备的故障风险大于所述风险下限的情况下，确定出所述目标数据获取方式为按照第二周期获取数据的第二获取方式，其中，所述第二周期短于所述第一周期。
  3. 如权利要求1或2所述的方法，其中，在所述利用目标决策模型对所述当前待分析数据进行分析，得到目标监测结果之后，所述方法还包括：
    在所述当前数据获取方式为所述第一获取方式，且所述目标监测结果指示所述目标设备的故障风险大于所述风险下限的情况下，生成指定风险事件，其中，所述指定风险事件用于指示所述目标设备的故障风险由小于或等于所述风险下限转变为大于所述风险下限；
    在当前获取方式为所述第一获取方式，所述更新后数据获取方式为所述第二获取方式的情况下，在所述按照所述更新后数据获取方式获取最新待分析数据之后，所述方法还包括：
    获取与指定时间段内获取的每个最新待分析数据对应的最新监测结果，其中，与所述每个最新待分析数据对应的最新监测结果是利用所述目标决策模型对所述最新待分析数据进行分析得到的；以及
    按照所述目标监测结果以及所述最新监测结果，确定出与所述指定风险事件 对应的风险控制策略。
  4. 如权利要求3所述的方法，其中，在所述获取与指定时间段内获取的每个最新待分析数据对应的最新监测结果之后，所述方法还包括：
    将所述目标监测结果以及所述最新监测结果，作为所述指定风险事件的算法评价结果，并将对应于所述目标监测结果以及所述最新监测结果的所有目标待分析数据，作为所述指定风险事件的原始设备数据，存储至历史风险事件库中，其中，所述目标待分析数据包括所述当前待分析数据和所述最新待分析数据；以及
    获取目标对象对所述指定风险事件进行评价得到的指定评价结果，并将所述指定评价结果作为所述指定风险事件的实际评价结果，存储至所述历史风险事件库，其中，所述实际评价结果用于指示所对应的风险事件实际认定的故障风险。
  5. 如权利要求4所述的方法，其中，所述方法还包括：
    从所述历史风险事件库中获取多个历史事件数据，其中，每个历史事件数据唯一对应于一个历史风险事件，每个历史风险事件对应的历史事件数据包括与所述每个历史风险事件对应的原始设备数据、算法评价结果以及实际评价结果；
    按照每个所述历史事件数据中的算法评价结果和实际评价结果，确定出所述目标决策模型的当前准确度；以及
    在所述目标决策模型的当前准确度低于预设阈值的情况下，根据所述多个历史事件数据对所述目标决策模型进行更新，得到更新后的目标决策模型。
  6. 如权利要求5所述的方法，其中，所述根据所述多个历史事件数据对所述目标决策模型进行更新，得到更新后的目标决策模型，包括：
    对于每个所述历史事件数据对应的历史风险事件，按照所述历史事件数据中的实际评价结果，对所述的历史事件数据中的算法评价结果进行修正，得到所述历史风险事件的修正评价结果；
    基于每个所述历史风险事件对应的原始设备数据、修正评价结果以及预设准确率，对所述目标决策模型中的黑盒分析模型进行重训练，得到更新后黑盒分析模型，其中，所述黑盒分析模型基于神经网络的故障预测模型，用于对待分析数据进行分析得到黑盒分析结果，所述更新后黑盒分析模型对待分析数据进行分析得到的黑盒分析结果的准确率不低于所述预设准确率；
    基于每个所述历史风险事件对应的原始设备数据、修正评价结果以及指定准确率，调整白盒分析模型中故障识别规则的判断阈值，得到更新后白盒分析模型，其中，所述白盒分析模型是所述目标决策模型中基于故障识别规则的模型，用于根据所述故障识别规则对待分析数据进行分析得到白盒分析结果，所述更新后白盒分析模型对待分析数据进行分析得到的白盒分析结果的准确率不低于所述指定准确率；以及
    基于所述目标决策模型中的目标决策规则、所述更新后黑盒分析模型以及所 述更新后白盒分析模型，得到所述更新后的目标决策模型，其中，所述目标决策规则用于根据目标决策模型中每个分析模型对数据的分析结果确定出监测结果。
  7. 如权利要求1至7中任一权利要求所述的方法，其中，所述目标监测类型为故障风险监测，所述利用目标决策模型对所述当前待分析数据进行分析，得到目标监测结果，包括：
    通过白盒分析模型对所述当前待分析数据进行分析，得到第一分析结果，其中，所述白盒分析模型是基于所述目标设备的故障识别规则的模型，所述目标决策模型包括所述白盒分析模型；
    通过黑盒分析模型对所述当前待分析数据进行分析，得到第二分析结果，其中，所述黑盒分析模型是基于神经网络的故障预测模型，所述目标决策模型包括所述黑盒分析模型；以及
    基于所述第一分析结果和所述第二分析结果，以及目标决策规则，得到与所述当前待分析数据对应的目标监测结果，其中，所述目标决策模型包括所述目标决策规则，所述目标决策规则用于根据所述目标决策模型中每个分析模型对数据的分析结果确定出监测结果，所述监测结果用于指示所述目标设备的故障风险。
  8. 设备监测系统，其包括采集终端和算法决策端：
    所述采集终端，配置为对目标设备采集得到目标数据类型的待分析数据；以及
    所述算法决策端，配置为按照当前数据获取方式，从所述采集终端获取当前待分析数据，其中，所述当前待分析数据是所述采集终端对所述目标设备采集得到的目标数据类型的数据；利用目标决策模型对所述当前待分析数据进行分析，得到目标监测结果，其中，所述目标决策模型用于对所述目标数据类型的数据进行分析得到目标监测类型的监测结果；确定出与所述目标监测结果对应的目标数据获取方式，通过所述目标数据获取方式对所述当前数据获取方式进行更新，得到更新后数据获取方式；按照所述更新后数据获取方式从所述采集终端获取最新待分析数据，并继续对所述目标设备进行监测。
  9. 电子设备，其包括：处理器、通信接口、存储器和通信总线，其中，所述处理器、所述通信接口和所述存储器通过所述通信总线完成相互间的通信；
    所述存储器，配置为存放计算机程序；
    所述处理器，配置为执行所述计算机程序时，实现权利要求1至7中任一权利要求所述的方法。
  10. 计算机可读存储介质，其中，所述存储介质包括存储的程序，其中，所述程序运行时执行上述权利要求1至7中任一权利要求所述的方法。
PCT/CN2022/133182 2022-09-30 2022-11-21 设备监测方法和系统、电子设备、存储介质 WO2024065988A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211215947.5A CN115617606A (zh) 2022-09-30 2022-09-30 设备监测方法和系统、电子设备、存储介质
CN202211215947.5 2022-09-30

Publications (1)

Publication Number Publication Date
WO2024065988A1 true WO2024065988A1 (zh) 2024-04-04

Family

ID=84860793

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/133182 WO2024065988A1 (zh) 2022-09-30 2022-11-21 设备监测方法和系统、电子设备、存储介质

Country Status (2)

Country Link
CN (1) CN115617606A (zh)
WO (1) WO2024065988A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116541241B (zh) * 2023-05-06 2024-03-15 华东医院 基于大数据的术后便携式可穿戴设备运行效率分析系统
CN117236704B (zh) * 2023-11-16 2024-02-06 中钢集团武汉安全环保研究院有限公司 一种钢铁生产线区域动态安全风险评估量化方法和装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105795497A (zh) * 2016-04-11 2016-07-27 浙江中烟工业有限责任公司 一种提高烘丝机故障分析的数据采集方法
CN111160685A (zh) * 2019-09-23 2020-05-15 上海安恪企业管理咨询有限公司 一种基于设备综合健康状况分析和管理的维修决策方法
WO2020211109A1 (zh) * 2019-04-17 2020-10-22 北京天泽智云科技有限公司 基于模型全生命周期管理的设备故障预警系统
CN112907911A (zh) * 2021-01-19 2021-06-04 安徽数分智能科技有限公司 一种基于设备工艺数据的智能异常识别与报警算法
CN114095519A (zh) * 2020-07-19 2022-02-25 智强通达科技（北京）有限公司 油库物联网设备状态监测及自动切换方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105795497A (zh) * 2016-04-11 2016-07-27 浙江中烟工业有限责任公司 一种提高烘丝机故障分析的数据采集方法
WO2020211109A1 (zh) * 2019-04-17 2020-10-22 北京天泽智云科技有限公司 基于模型全生命周期管理的设备故障预警系统
CN111160685A (zh) * 2019-09-23 2020-05-15 上海安恪企业管理咨询有限公司 一种基于设备综合健康状况分析和管理的维修决策方法
CN114095519A (zh) * 2020-07-19 2022-02-25 智强通达科技（北京）有限公司 油库物联网设备状态监测及自动切换方法
CN112907911A (zh) * 2021-01-19 2021-06-04 安徽数分智能科技有限公司 一种基于设备工艺数据的智能异常识别与报警算法

Also Published As

Publication number Publication date
CN115617606A (zh) 2023-01-17

Similar Documents

Publication Publication Date Title
WO2024065988A1 (zh) 设备监测方法和系统、电子设备、存储介质
EP3902992B1 (en) Scalable system and engine for forecasting wind turbine failure
US20220245297A1 (en) Scalable system and method for forecasting wind turbine failure using scada alarm and event logs
EP3557819B1 (en) Server failure detection method and system
US11403164B2 (en) Method and device for determining a performance indicator value for predicting anomalies in a computing infrastructure from values of performance indicators
US20230152795A1 (en) Scalable systems and methods for assessing healthy condition scores in renewable asset management
CN112188531B (zh) 异常检测方法、装置、电子设备及计算机存储介质
US11307916B2 (en) Method and device for determining an estimated time before a technical incident in a computing infrastructure from values of performance indicators
CN117176560B (zh) 一种基于物联网的监测设备监管系统及其方法
CN112862012A (zh) 一种基于lstm模型的运维系统异常预警方法、装置及设备
CN102032115A (zh) 用于分析报告数据的系统和方法
WO2021042687A1 (zh) 提升预测性维护模型适应能力的方法及装置
WO2020242553A1 (en) Failure prediction using gradient-based sensor identification
RU2686257C1 (ru) Способ и система удалённой идентификации и прогнозирования развития зарождающихся дефектов объектов
US11675643B2 (en) Method and device for determining a technical incident risk value in a computing infrastructure from performance indicator values
CN111382494A (zh) 用于检测工业机器的感知数据中的异常的系统及方法
CN112270429A (zh) 基于云边协同的动力电池极片制造设备维护方法和系统
US20180307218A1 (en) System and method for allocating machine behavioral models
CN116308304A (zh) 基于元学习概念漂移检测的新能源智慧运维方法及系统
CN116307641A (zh) 一种面向数字化电厂的资源协同调度管理方法及系统
Mazidi et al. A performance and maintenance evaluation framework for wind turbines
CN104863798A (zh) 基于分布式存储的风电机组状态监测系统及其监测方法
CN115356990A (zh) 基于深度学习的料场设备故障预测方法、系统及存储介质
CN117076260B (zh) 一种参数及设备异常检测方法及装置
CN117473445B (zh) 基于极限学习机的设备异常分析方法及装置