KR20220168269A - System and method for predicting and analyzing trouble of mechanical equipment using federated learning - Google Patents

Abstract

The present invention relates to a mechanical equipment failure predictive analysis system and method using federated learning implemented to enable real-time predictive analysis of mechanical equipment failures based on artificial intelligence (AI) using federated learning. A data collector collects IoT data from mechanical equipment. A federated learner receives the IoT data collected from the data collector and performs federated learning to generate federated learning data. An AI predictive analyzer receives the federated learning data generated by the federated learner and performs predictive analysis of mechanical equipment failures in real time based on AI. Accordingly, real-time mechanical equipment control and analysis are possible.

Description

System and method for predicting and analyzing trouble of mechanical equipment using federated learning}

The technical field of the present invention relates to a mechanical facility failure predictive analysis system and method using federated learning. It relates to a hardware failure prediction analysis system and method using federated learning implemented so as to be possible.

Management of machinery and equipment is a general activity to maximize the function of machinery and equipment by planning, maintaining, and improving equipment in accordance with company policy in order to increase productivity, reduce opportunity loss, and improve profitability. In addition, it can include activities to improve production efficiency and quality and reduce maintenance costs by maintaining the entire process from birth to destruction of machinery and equipment in a perfect state at all times.

Looking at the definition of maintenance and maintenance for machinery, maintenance can be defined as a maintenance activity that maintains machinery in its original state, and similar improvement of machinery refers to a state that improves its original performance. Accordingly, maintenance has the meaning of preservation, and maintenance mobilizes people from all fields related to the asset to preserve the function of the physical asset, manages the failure or the result of failure, and applies the least cost and most efficient technology. It can be defined as satisfying the users who use the asset.

As mechanization and automation are applied to machinery (or devices or systems), and the share of machinery in corporate management is increasing, product quality, cost, production volume, etc. and reliability are becoming more frequent. In addition, the environment around companies is changing into a form that can survive only when it meets the needs of consumers who demand urgency and diversity at the same time. Accordingly, how much flexibility and reliability can be obtained in terms of mechanical facilities is a source of competitiveness for companies, and maintenance and improvement of mechanical facilities occupy an important position in the manufacturing and production process.

Korean Patent Registration No. 10-1713985 (registered on March 2, 2017) predicts failures of the device in advance and secures a maintenance plan method for the device that can maintain the performance of the device, thereby improving the operation performance and reliability of the device. In the meantime, a predictive maintenance method and apparatus are disclosed that can reduce the maintenance cost of equipment by maintaining the life of the equipment and performing efficient maintenance activities. According to the disclosed technology, a predictive maintenance method includes the steps of a predictive maintenance device collecting maintenance data from a device subject to maintenance; Classifying, by the predictive maintenance device, maintenance data into normal operation data and failure occurrence data; generating, by the predictive maintenance device, device failure determination criteria based on normal operation data and failure occurrence data; and predicting, by the predictive maintenance device, a failure of the device to be maintained based on a device failure determination criterion and new input data received from the device to be maintained.

Korean Registered Patent No. 10-1596004 (registered on Feb. 15, 201) discloses an intelligent building facility management system and a management method therefor. In the intelligent building facility management system, a database; manager interface device; It is connected to the facilities provided in the building, receives registration information for each facility and stores it in the database, creates virtual facility information corresponding to the facilities and stores it in the database, and stores the virtual facility information or virtual facility information. By selectively mashing up the output values of the facilities corresponding to the facilities at the request of the manager through the manager interface device, complex virtual facility information is created and stored in the database, and whenever a specific event occurs, status information about the facilities is collected, and It is characterized by having a service management device that calculates the efficiency of facilities belonging to virtual facility information or complex virtual facility information based on status information, and provides predetermined facility operation control information corresponding to the calculated efficiency to facilities. to be According to the disclosed technology, a manager directly and selectively mash-ups virtual facilities corresponding to various facilities provided in a building to create meaningful new complex virtual facilities, as well as building efficiency based on these new complex virtual facilities. can be judged or controlled.

In the conventional technology as described above, in the case of a facility management system, it is applied to the production facility and process field, and the presence or absence of abnormality of the production facility and process facility installed in a factory, etc., monitoring and diagnosis management of operational performance are performed, and most of the abnormal conditions are analyzed. For this purpose, data processing was processed in the system server and then diagnosed and notified. In the case of the existing DDC, only data input/output and control functions were performed, and the predictive analysis function used in most BEMS/FEMS is a The predictive analysis function for facilities was insufficient.

Korean Patent Registration No. 10-1713985 Korean Patent Registration No. 10-1596004

The problem to be solved by the present invention is to solve the above-mentioned disadvantages, implemented so that failures of mechanical equipment can be analyzed in real time based on AI (artificial intelligence) using federated learning. It is to provide a hardware failure predictive analysis system and method using federated learning.

As a means for solving the above problems, according to one feature of the present invention, a data collector for collecting IoT data from machinery; A federated learner for receiving the IoT data collected by the data collector and performing federated learning to generate federated learning data; And it provides a hardware failure predictive analysis system using federated learning including an AI predictive analyzer for receiving the federated learning data generated by the federated learner and predictively analyzing the failure of machinery in real time based on AI.

In one embodiment, the data collector is characterized by collecting IoT data measured or sensed by a measuring device or sensor installed in an automatic control facility.

In one embodiment, the data collector, in the case of flow data measurement in the IoT complex valve, executes a 100% opening degree command through the control device and checks three times whether the reference instantaneous flow rate value in the flow test device is monitored, and then calculates the average accuracy It is characterized by calculating.

In one embodiment, the data collector, in the case of measuring pressure data in the IoT complex valve, sets the condition of the test device to 0.8 MPa and checks 5 times whether the pressure value is monitored with the display unit connected to the valve, and then calculates the average accuracy. It is characterized by calculating.

In one embodiment, the data collector, in the case of temperature data measurement in the IoT complex valve, sets a specific temperature in the chamber, maintains the valve in a stable state for 2 hours, and checks 5 times whether the temperature value is monitored by the display unit. It is characterized in that the average accuracy is calculated.

In one embodiment, the data collector, in the case of smart DDC input measurement device data monitoring, is characterized in that it calculates average accuracy after checking whether data input values of heterogeneous measurement facilities are monitored five or more times.

In one embodiment, in the case of the IoT measurement data collection and storage algorithm, the data collector selects 5 types of data to be collected and 5 types of data to be stored from the IoT sensor device and tests the accuracy of the algorithm.

In one embodiment, in the case of the predictive maintenance algorithm, the data collector measures an accuracy error rate of the predictive maintenance algorithm using test data of 10 types of failures.

In one embodiment, in the case of the federated learning-based predictive maintenance algorithm, the data collector measures the processing speed of the predictive maintenance algorithm using 10 types of test data.

In one embodiment, the data collector, in the case of an integrated solution for predictive maintenance, uses 10 types of test data to test the stability of whether the composite solution for predictive maintenance service performs prediction without errors. to be

(L/min) 도달 여부를 확인하고, IoT 복합밸브 내 제어장치의 표시부를 통해 해당 유량 값이 모니터링 되는지를 확인하고, 이것을 3회 반복 시행 후 각 회별 기준기와 밸브의 순간유량 값의 평균 오차 값을 산출하여 95% 이상 정확도를 갖는지를 시험하는 것을 특징으로 한다.In one embodiment, the data collector, in the case of measuring the flow rate data in the IoT complex valve, after executing 100% of the opening through the control device connected to the IoT complex valve, the instantaneous flow rate value of the standard in the medium-sized flow test device 39

(L/min), check whether the corresponding flow rate value is monitored through the display of the control device in the IoT complex valve, and after repeating this three times, the average error value of the instantaneous flow rate value of the standard and valve for each time It is characterized by calculating and testing whether it has an accuracy of 95% or more.

In one embodiment, the data collector, in the case of measuring pressure data in the IoT complex valve, checks whether the pressure in the IoT complex valve is monitored by the display unit of the control device, and sets the conditions of the test device to 20 ° C and 0.8 MPa. to confirm that the pressure value is accurately measured with the display of the control device, set the condition of the test device to 20°C, 1.6 MPa, and check whether it is monitored with the display, and repeat this 5 times, and then the measured pressure value is the standard It is characterized by testing the accuracy measured within the ± 0.5% error range of the pressure value.

In one embodiment, the data collector, in the case of measuring temperature data in the IoT composite valve, set the IoT composite valve and the test device environment to (0.0 ± 3.0) ° C and maintain the IoT composite valve in a stable state for 1 hour , Check whether it is monitored by the display of the control device, set the temperature of the test device to (40±3) °C, keep the IoT complex valve in a stable state for 2 hours, check the measured value, and repeat this 5 times After implementation, it is characterized in that the accuracy is calculated after checking whether the temperature measurement value of the display unit maintains within the ± 1.0 ° C error range.

In one embodiment, the data collector, in the case of smart DDC input measurement device data monitoring, transmits the data detected by the instrument to the smart DDC through a communication network, and determines whether the data input value is monitored by the instrument connected to the smart DDC. After confirming, the data transmitted to the smart DDC is recorded on the smart DDC side, and this data is stored in real time or by using a sensor-based simulation tool, and after checking more than 5 times whether the data input value is monitored It is characterized in that the average accuracy is calculated.

In one embodiment, in the case of the IoT measurement data collection and storage algorithm, after executing the IoT measurement data collection algorithm, the data collector tests whether 5 types of actual data are normally collected from the IoT complex valve, and repeats the test 10 times. It is characterized in that the accuracy of the algorithm is tested by executing.

In one embodiment, the data collector, in the case of a predictive maintenance algorithm, after executing a federated learning-based predictive maintenance algorithm, training the collected and stored IoT measurement data and existing maintenance data to build a predictive model, After confirming the derivation of the prediction results for the test data of 10 types of failures, it is characterized by measuring the average error rate after checking the error rate (CvRMSE) of the prediction model at least 5 times using 30% of the total data.

In one embodiment, the training of the IoT measurement data and the existing maintenance data, in the case of the IoT complex valve, checks whether to respond according to the data transmission and reception cycle and whether to update the data, and in the case of the control drive unit, the current state of the facility is returned It is characterized by checking the control status through matching of signals and control signals, and in the case of smart DDC, checking whether communication is transmitted or received at regular intervals in connection with the supervisory control monitoring system.

In one embodiment, the data collector, in the case of a federated learning-based predictive maintenance algorithm, after executing the federated learning-based predictive maintenance algorithm, trains the collected and stored IoT measurement data to build a predictive model, and then builds a predictive model. It is characterized in that the prediction time for the test data of is measured.

In one embodiment, the data collector, in the case of an integrated solution for predictive maintenance, executes an integrated predictive maintenance solution using IoT measurement data, and then confirms the derivation of predictive results for 10 types of test data, It is characterized by testing the stability of the software through repeated tests 10 times.

In one embodiment, the data collector is characterized in that the equipment failure data is classified into measurement signal error, communication state error, reverse flow error, input sensor error, and inoperability.

In one embodiment, the data collector is characterized by utilizing Apache Flume for real-time data collection, HDFS, HBase, Impala for managing big data in a distributed environment, and Apache Spark for real-time data analysis and processing.

In one embodiment, the federated learner applies a federated learning technique, so that a server and a plurality of edge devices linked to the server distribute roles to each other to receive IoT data collected by the data collector and learn. .

In one embodiment, the federated learner builds a federated learning system including a server and a plurality of DDCs in order to apply federated learning techniques, and mounts a processor capable of storing and calculating IoT data in the DDC to generate IoT data. It is characterized in that it performs associative learning for.

In one embodiment, the DDC is characterized in that it is capable of storing and processing data, and controls field devices with digital values directly at a specific location by applying federated learning based on this.

In one embodiment, the federated learner performs preprocessing analysis and control at the edge device level, but uses federated learning and applies distributed processing edge computing technology by the edge device to transmit the IoT data collected by the data collector It is characterized by receiving and learning pre-processing analysis.

In one embodiment, the federated learner is characterized in that it receives the IoT data collected by the data collector and learns primarily through preprocessing analysis.

In one embodiment, the federated learner is characterized in that it receives a plurality of sensing information from the data collector, preprocesses and analyzes them first, and compares them with other sensing values in real time to determine whether it is a temporary failure or a failure.

In one embodiment, the DDC is characterized in that, while primarily diagnosing a failure of a mechanical facility, it performs a failure diagnosis by comparing collected data with learned data of a server.

In one embodiment, the federated learner is characterized in that common conditions for failure prediction analysis are classified through federated learning to find a hardware common use analysis model.

In one embodiment, the AI predictive analyzer performs real-time failure prediction and management of mechanical facilities by applying AI techniques to automatic control facility management, and detects, analyzes, and controls and manages abnormal occurrences of mechanical facilities in real time. to be characterized

In one embodiment, the AI predictive analyzer is characterized by applying analysis techniques of logic, decision-making, rules, and models to federated learning data generated by the federated learner.

In one embodiment, the AI predictive analyzer is characterized in that it creates a rule corresponding thereto by using the primary pre-processed and analyzed learning data generated by the federated learner.

In one embodiment, the AI predictive analyzer is characterized in that it remodels a predefined rule using primarily pre-processed and analyzed learning data.

In one embodiment, the AI predictive analyzer is characterized in providing a prediction service by modeling through analysis of primarily pre-processed and analyzed learning data.

In one embodiment, the AI predictive analyzer regularly checks the state of machinery with AI-based CBM using the primarily pre-processed and analyzed learning data transmitted from the associated learner to identify the precursor of failure, and finally It is characterized by performing predictive maintenance before reaching failure.

In one embodiment, the AI predictive analyzer is characterized by modeling and storing preliminary analysis and judgment criterion data for failures of mechanical facilities and A/S situations and storing them in memory.

In one embodiment, the AI predictive analyzer is characterized in that it provides a multi-adjusted visualization of the analysis data to the user.

In one embodiment, the AI predictive analyzer is characterized in dividing predictive maintenance into daily inspection and monitoring when servicing predictive maintenance.

In one embodiment, the AI predictive analyzer distinguishes whether the collected data is judgment data or quantitative data during daily inspection, and in the case of judgment data, diagnoses whether it is abnormal or normal using data determined by visual observation by an observer, In the case of weighing data, it is characterized in that the diagnosis is made based on the standard of the upper and lower limit specifications set in advance using data collected through a measuring device or sensor.

In one embodiment, the AI predictive analyzer, during monitoring, distinguishes whether the collected data is digital data or analog data, and in the case of digital data, collects the state signal of the measuring device or sensor to diagnose whether it is abnormal or normal, In the case of analog data, it is characterized in that measurement values of measuring devices or sensors are collected and diagnosed based on pre-set upper and lower specification limit lines or diagnosis based on management limit lines using control chart patterns.

In one embodiment, the AI predictive analyzer performs predictive analysis according to the type of failure of the automatic control equipment, and in the case of a failure of the IoT complex valve data measurement device, checks whether the response and data update are determined according to the data transmission/reception cycle, and IoT In the case of a failure of the control drive unit of the complex valve, the control status is checked through whether the return signal of the current state and the control signal match from the facility. It is characterized by confirming.

In one embodiment, in order to predict and determine the failure situation of the building's air conditioning and heating system, the AI predictive analyzer classifies it as a need for maintenance when the measurement value shows an error of more than a certain level from the average measurement of other buildings, and When the above emergency state information is collected, it is characterized in that it is classified as a need for full-scale system repair.

As a means for solving the above problems, according to another feature of the present invention, the step of the data collector collecting IoT data from the machine equipment; A step in which the federated learner receives the IoT data collected by the data collector and performs federated learning to generate federated learning data; and the AI predictive analyzer receiving the federated learning data generated by the federated learner and performing a real-time predictive analysis of machine failure based on AI.

As an effect of the present invention, a mechanical facility failure predictive analysis system and method using federated learning implemented to perform predictive analysis of mechanical facility failure in real time based on AI (artificial intelligence) using federated learning. By providing it, it is possible to perform failure prediction analysis of automatic control equipment such as valves used for air conditioning and heating by applying failure prediction analysis technology to operating automatic control facilities such as buildings and factories, and AI-based smart DDC using federated learning. Real-time control and analysis of mechanical equipment is possible through real-time, predictive analysis of abnormal conditions of mechanical equipment such as air conditioning, air conditioning, and automatic control equipment used in the operation of buildings and factories, and data processing analysis and control in real time through federated learning is possible, and thus the control response speed can be improved by improving the response speed for control and management of mechanical facilities while reducing the system load.

1 and 2 are diagrams illustrating a hardware failure prediction and analysis system using federated learning according to an embodiment of the present invention.
3 is a diagram illustrating an example of a predictive maintenance service of the AI predictive analyzer in FIG. 1 .
4 is a diagram illustrating a hardware failure predictive analysis method using federated learning according to an embodiment of the present invention.

Hereinafter, with reference to the accompanying drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily carry out the present invention. However, since the description of the present invention is only an embodiment for structural or functional description, the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, since the embodiment can be changed in various ways and can have various forms, it should be understood that the scope of the present invention includes equivalents capable of realizing the technical idea. In addition, since the object or effect presented in the present invention does not mean that a specific embodiment should include all of them or only such effects, the scope of the present invention should not be construed as being limited thereto.

The meaning of terms described in the present invention should be understood as follows.

Terms such as "first" and "second" are used to distinguish one component from another, and the scope of rights should not be limited by these terms. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element. It should be understood that when an element is referred to as “connected” to another element, it may be directly connected to the other element, but other elements may exist in the middle. On the other hand, when an element is referred to as being “directly connected” to another element, it should be understood that no intervening elements exist. Meanwhile, other expressions describing the relationship between components, such as “between” and “immediately between” or “adjacent to” and “directly adjacent to” should be interpreted similarly.

Singular expressions should be understood to include plural expressions unless the context clearly dictates otherwise, and terms such as “comprise” or “having” refer to a described feature, number, step, operation, component, part, or It should be understood that it is intended to indicate that a combination exists, and does not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs, unless defined otherwise. Terms defined in commonly used dictionaries should be interpreted as consistent with meanings in the context of related art, and cannot be interpreted as having ideal or excessively formal meanings unless explicitly defined in the present invention.

Now, a hardware failure prediction analysis system and method using federated learning according to an embodiment of the present invention will be described in detail with reference to the drawings.

1 and 2 are diagrams illustrating a hardware failure predictive analysis system using federated learning according to an embodiment of the present invention, and FIG. 3 illustrates an example of a predictive maintenance service of the AI predictive analyzer in FIG. 1 it is a drawing

1 to 3, the hardware failure prediction analysis system 100 using federated learning includes a plurality of data collectors 110, a federated learner 120, and an AI predictive analyzer 130.

The data collector 110 collects IoT data from the hardware and delivers it to the federated learner 120.

In one embodiment, the data collector 110 is a measuring device or sensor (eg, a thermometer, a flowmeter, etc.) installed in an automatic control facility (eg, air-conditioning facility, IoT complex valve facility, etc.) , pressure gauge, etc.) can collect IoT data from (111).

In one embodiment, the data collector 110, in the case of measuring the flow rate data in the IoT complex valve, executes a 100% opening command through the control device and checks three times whether the standard instantaneous flow rate value in the flow test device is monitored and averages Accuracy can be calculated, and in the case of pressure data measurement in the IoT complex valve, the average accuracy can be calculated after setting the condition of the test device to 0.8 MPa and checking whether the pressure value is monitored with the display connected to the valve 5 times. In the case of temperature data measurement in the IoT complex valve, after setting a specific temperature in the chamber and maintaining the valve in a stable state for 2 hours, the average accuracy can be calculated after checking whether the temperature value is monitored by the display unit 5 times. In the case of smart DDC input measurement device data monitoring, average accuracy can be calculated after checking whether the data input values of heterogeneous measurement facilities (eg, power, air conditioning, etc.) are monitored more than 5 times, and IoT measurement data In the case of the collection and storage algorithm, the accuracy of the algorithm can be tested by selecting 5 types of data to be collected and 5 types to be stored from the IoT sensor device. Flow, pressure, temperature data measurement device failure, control drive failure, smart DDC communication transmission failure, etc.) can be used to measure the accuracy error rate (CvRMSE) of the predictive maintenance algorithm using 10 types of test data, and predictive maintenance based on federated learning For the maintenance algorithm, the processing speed of the predictive maintenance algorithm can be measured using 10 types of test data, and in the case of an integrated solution for predictive maintenance, 10 types of test data can be used to measure the processing speed The stability of the composite solution making predictions without error can be tested.

(L/min) 도달 여부를 확인하고, IoT 복합밸브 내 제어장치의 표시부를 통해 해당 유량 값이 모니터링 되는지를 확인하고, 이것을 3회 반복 시행 후 각 회별 기준기와 밸브의 순간유량 값의 평균 오차 값을 산출하여 95% 이상 정확도를 갖는지를 시험할 수 있다.In one embodiment, the data collector 110, in the case of measuring the flow rate data in the IoT complex valve, executes 100% of the opening through the control device connected to the IoT complex valve, and the instantaneous flow rate value of the standard in the medium-sized flow test device 39

(L/min), check whether the corresponding flow rate value is monitored through the display of the control device in the IoT complex valve, and after repeating this three times, the average error value of the instantaneous flow rate value of the standard and valve for each time You can test whether it has more than 95% accuracy by calculating .

In one embodiment, the data collector 110, in the case of measuring pressure data in the IoT complex valve, checks whether the pressure in the IoT complex valve is monitored by the display unit of the control device, and sets the conditions of the test device to 20 ° C and 0.8 MPa. Set to , check whether the pressure value is accurately measured by the display of the control device, set the condition of the test device to 20°C, 1.6 MPa to check if it is monitored by the display, repeat this 5 times, and then measure the pressure value The accuracy measured within the ±0.5% error range of this reference pressure value can be tested.

In one embodiment, the data collector 110, in the case of measuring temperature data in the IoT composite valve, sets the IoT composite valve and the test device environment to (0.0 ± 3.0) ° C and puts the IoT composite valve in a stable state for 1 hour. and check whether it is monitored by the display of the control device, set the temperature of the test device to (40±3) °C, and after maintaining the IoT complex valve in a stable state for 2 hours, check the measured value, and set it to 5 Accuracy can be calculated after confirming that the temperature measurement value of the display part maintains within the ±1.0°C error range after repeated execution.

In one embodiment, the data collector 110, in the case of smart DDC input measurement device data monitoring, transmits the data detected by the building facility (power, air conditioning, etc.) meter to the smart DDC through a communication network, and is connected to the smart DDC It is checked whether the data input value is monitored by the building facility (electric power, air conditioning, etc.) meter, and the data transmitted to the smart DDC is recorded on the smart DDC side, and this data is stored in real time or utilized by a sensor-based simulation tool. After saving the data and checking more than 5 times whether the data input value is monitored, the average accuracy can be calculated.

In one embodiment, the data collector 110, in the case of the IoT measurement data collection and storage algorithm, after executing the IoT measurement data collection algorithm, tests whether 5 types of actual data are normally collected from the IoT complex valve, and 10 times You can test the accuracy of the algorithm by running iterative tests.

In one embodiment, the data collector 110, in the case of the predictive maintenance algorithm, after executing the federated learning-based predictive maintenance algorithm, collected and stored IoT measurement data and existing maintenance data (eg, IoT complex valve In the case of the control drive unit, the control status is checked by checking whether the response signal of the current state and the control signal match in the case of the control drive unit, and in the case of smart DDC After establishing a predictive model by training (e.g., checking whether communication is transmitted/received at regular intervals (1 second) in connection with the supervisory control monitoring system), the type of failure (e.g., IoT complex valve flow rate, pressure, temperature data) Failure of measurement device, failure of control drive unit, failure of smart DDC communication transmission, etc.) After confirming the derivation of prediction results for 10 types of test data, using 30% of the total data, after checking the error rate (CvRMSE) of the prediction model more than 5 times The average error rate can be measured.

In one embodiment, the data collector 110, in the case of a federated learning-based predictive maintenance algorithm, after executing the federated learning-based predictive maintenance algorithm, trains the collected and stored IoT measurement data to build a predictive model, Prediction times can be measured for 10 types of test data.

In one embodiment, in the case of an integrated solution for predictive maintenance, the data collector 110 confirms the derivation of predictive results for 10 types of test data after executing the predictive maintenance integrated solution using IoT measurement data. And the stability of the software can be tested by repeating this test 10 times.

In one embodiment, the data collector 110 may classify the machine facility (automatic control) failure data as measurement signal error, communication status error, reverse flow error, input sensor error, inoperability, and the like.

In one embodiment, the data collector 110 may collect and store real-time data from an IoT device (eg, an IoT complex valve), and real-time flow rate, temperature, and pressure data from a flow rate, temperature, and pressure sensor of the IoT complex valve. and Apache Flume for real-time data collection, HDFS, HBase, Impala for managing big data in a distributed environment, and Apache Spark for real-time data analysis and processing. can be utilized

The federated learner 120 performs federated learning on the IoT data transmitted from the data collector 110 to generate federated learning data and passes it to the AI predictive analyzer 130.

In one embodiment, the federated learner 120 applies a federated learning technique, which is a machine learning technique, so that the server 121 and a plurality of edge devices linked to the server 121 play a role instead of learning on a single server. It is possible to learn the IoT data transmitted from the data collector 110 by distributing.

In one embodiment, the federated learner 120 builds a federated learning system including a server 121 and a plurality of direct digital controllers (DDCs) 122 to apply the federated learning technique. In this case, federated learning for IoT data may be performed by mounting a processor capable of storing and calculating IoT data in the DDC 122. Here, the DDC 122 can directly control field devices with digital values (eg, values or communication) at a specific location (eg, a central monitoring panel, a disaster prevention room, etc.).

In one embodiment, the DDC 122 is capable of storing and processing data, and based on this, federated learning is applied to support fast control and calculation speed, so that the AI predictive analyzer 130 performs failure prediction in real time. allows you to

In one embodiment, the federated learner 120 may perform preprocessing analysis and control at the edge device level. At this time, the AI predictive analyzer 130 may use federated learning and apply distributed processing edge computing technology by the edge device. ), preprocessing analysis can be learned in the previous step.

In one embodiment, the federated learner 120 may learn the IoT data transmitted from the data collector 110 primarily through pre-processing analysis.

In one embodiment, the federated learner 120 primarily pre-processes and analyzes a plurality of sensing information transmitted from the data collector 110 and compares them with other sensing values in real time to determine whether it is a temporary failure or a failure. can

In one embodiment, the DDC 122 may primarily diagnose a failure of a mechanical facility, and at this time, the collected data may be compared with the learned data of the server 121 to perform a failure diagnosis.

In one embodiment, the federated learner 120 may classify common conditions for failure prediction analysis through federated learning to find a hardware common use analysis model.

The AI predictive analyzer 130 uses the federated learning data transmitted from the federated learner 120 to predictively analyze failures of machinery based on AI in real time.

In one embodiment, the AI predictive analyzer 130 can perform failure prediction and management of mechanical facilities in real time by applying AI techniques to automatic control facility management, detecting, analyzing, control can be managed.

In one embodiment, the AI predictive analyzer 130 may improve productivity and efficiency by applying analysis techniques such as logic, decision-making, rules, and models to federated learning data transmitted from the federated learner 120. there is.

In one embodiment, the AI predictive analyzer 130 uses the federated learning data transmitted from the federated learner 120 (i.e., primarily preprocessed-analyzed learning data) to apply the corresponding rule (i.e., the most suitable and efficient rules) can be created. At this time, pre-defined rules may be remodeled using the first pre-processed and analyzed learning data.

In one embodiment, the AI predictive analyzer 130 may provide a prediction service through modeling through analysis of primarily pre-processed and analyzed learning data, rather than deriving and predicting a pattern based on energy usage.

In one embodiment, the AI predictive analyzer 130 does not detect and analyze abnormal conditions using only predefined rules, but federated learning data transmitted from the federated learner 120 (ie, primarily preprocessed and analyzed Learning data), AI-based CBM (condition based maintenance) regularly checks the condition of machinery and equipment to identify the precursors of failure, so that predictive maintenance can be performed before reaching the final failure. there is.

In one embodiment, the AI predictive analyzer 130, even if the failure of the machine equipment has never occurred, the basic federation learning data for predicting the failure of the machine equipment as big data in advance by failure type database It may be stored in the memory, and the failure of the machine equipment may be predicted by comparing and analyzing the federated learning data transmitted from the federated learner 120 with the big data stored in the memory. New federated learning data can be continuously upgraded in memory to improve the reliability of big data.

In one embodiment, the AI prognostic analyzer 130 provides preliminary analysis and criterion data for failures of mechanical facilities and A/S situations (eg, IoT complex valves, facility control monitoring devices, digital instruments, etc.) It can also be modeled and stored in memory.

In one embodiment, the AI predictive analyzer 130 may provide a user with multiple coordinated visualization of the analysis data.

In one embodiment, the AI predictive analyzer 130, when servicing predictive maintenance, can be divided into daily inspection and monitoring, as shown in FIG. 3, and auto gatering during daily inspection When this is not possible, it is possible to distinguish whether the collected data is judgment data or weighing data, and in the case of judgment data, the observer can use the data judged by visual observation to diagnose whether it is abnormal or normal, and in the case of metering data, the measuring device It is possible to use the data collected through the sensor 111 to diagnose based on the standard of the upper and lower limit limits set in advance, and also, when monitoring, auto gating is possible, whether the collected data is digital data or analog data. can be distinguished, and in the case of digital data, the state signal of the measuring device or sensor 111 can be collected to diagnose whether it is 1 (i.e., abnormal) or 0 (i.e., normal), and in the case of analog data, measurement The measured values of the device or sensor 111 may be collected and diagnosed based on a pre-set upper and lower specification limit line, or may be diagnosed based on a management limit line using a control chart pattern.

In one embodiment, the AI predictive analyzer 130 is a type of automatic control facility failure (eg, IoT complex valve data measurement device failure (flow rate, pressure, temperature) cannot update measurement data due to failure, IoT complex valve control drive unit Inability to control flow rate and temperature due to failure, failure to monitor and control facility data due to failure of smart DDC communication transmission and reception, etc.), predictive analysis is performed. You can check whether there is a response according to the transmission/reception cycle (up to 30 seconds) and whether the data is updated. In the case of a failure of the IoT complex valve control drive unit, you can check the control status through the matching of the current status reply signal and control signal from the facility. , In case of smart DDC communication transmission/reception failure, it is possible to check communication transmission/reception at regular intervals (1 second) in connection with the monitoring control monitoring system.

In one embodiment, the AI prognosis analyzer 130, in order to predict and determine the failure situation of the building's air-conditioning system, has an error of more than a certain amount (eg, 5%) from the average measurement of the case of a building in which the measured values of the components are different. , it can be classified as maintenance required, when emergency state information above a certain level is collected (eg, about 30% of components), and total system repair is required.

The mechanical equipment failure prediction analysis system 100 using federated learning having the configuration described above is implemented to perform predictive analysis of mechanical failures in real time based on AI by applying federated learning, so that buildings, factories, etc. By applying failure prediction analysis technology to operating automatic control facilities, it is possible to perform failure prediction analysis of automatic control equipment devices such as valves used for air conditioning and heating, and real-time mechanical equipment through AI-based smart DDC (122) using federated learning Control and analysis are possible, predictive analysis of abnormal conditions of mechanical facilities such as air conditioning, air conditioning, and automatic control facilities used in the operation of buildings and factories, and data processing analysis and control are possible in real time through combined learning. Control response speed can be improved by improving response speed for machine facility control and management while reducing system load.

The hardware failure prediction analysis system 100 using federated learning having the configuration described above performs maintenance of machinery in accordance with the rapid development of advanced technologies, such as the Internet of Things (IoT), artificial intelligence (AI) and By converging 4th industrial technology such as data analysis with factories and machines, it improves the performance of mechanical facilities and maintains and manages them in the best condition. It can prevent huge economic losses to the company in advance. In addition, the hardware failure prediction analysis system 100 using federated learning having the above-described configuration can enhance corporate competitiveness through productivity improvement and cost reduction, through various measuring devices or sensors 111 IoT data can be collected, analyzed, and evaluated, and defects in mechanical facilities can be notified before failure, thereby reducing cost and maintenance costs due to failure and improving productivity. The disadvantages of the post-maintenance method can be improved, and maintenance can be carried out before mechanical equipment fails.

The mechanical facility failure prediction analysis system 100 using federated learning having the configuration described above can respond to the increase in mechanical facility management data required for the operation of buildings, factories, etc. Minimization of downtime by improving the response speed for mechanical facility control and management, as the government certification system such as BEMS and ZEB (zero energy building) has become mandatory, and the control and management data of mechanical facilities has increased significantly due to the increase in certification level. and reduce restoration costs. In addition, the machine facility failure prediction analysis system 100 using federated learning having the configuration described above can perform failure prediction analysis and control for operating facilities according to the requirements of the customer, but only diagnoses and analyzes the production facilities. It is not a failure prediction analysis system that provides, but a failure prediction analysis system that can monitor and control the presence of abnormalities and operational performance of mechanical equipment used in the operation of buildings and factories. Since is fatal, it is possible to predict and respond before failure occurs. For example, in the case of a semiconductor factory, the operation of the factory environment, such as constant temperature and humidity, is also very important, and great damage occurs in the event of an abnormality in mechanical facilities such as air-conditioning and air-conditioning facilities. , it is possible to perform predictive management instead of post-processing.

The hardware failure prediction analysis system 100 using federated learning having the configuration described above performs AI-based real-time failure prediction analysis for predictive maintenance, using an artificial intelligence-based smart DDC 122 capable of federated learning. While using it, build a big data platform by converging IoT, big data, and edge computing technology with automatic control equipment, and identify the possibility of failure through AI-based predictive maintenance and respond in advance to prevent downtime, system server You can perform capacity reduction, data management, and more.

4 is a diagram illustrating a hardware failure predictive analysis method using federated learning according to an embodiment of the present invention.

Referring to FIG. 4, the data collector 110 collects IoT data from the hardware and delivers it to the federated learner 120 (S401).

In transmitting the IoT data in the above-described step S401, measuring devices or sensors (eg, thermometers, flowmeters, etc.) installed in automatic control facilities (eg, air-conditioning facilities, IoT complex valve facilities, etc.) In the data collector 110 having a pressure gauge) 111, the corresponding measuring device or sensor measures or detects the state of the automatic control facility, collects the detected IoT data, and transmits it to the associated learner 120 can give

In transmitting the IoT data in the above-described step S401, in the case of measuring the flow rate data in the IoT complex valve, the data collector 110 executes a 100% opening command through the control device and monitors the reference instantaneous flow rate value in the flow test device. The average accuracy can be calculated after checking three times whether it is correct.

(L/min) 도달 여부를 확인하고, IoT 복합밸브 내 제어장치의 표시부를 통해 해당 유량 값이 모니터링 되는지를 확인하고, 이것을 3회 반복 시행 후 각 회별 기준기와 밸브의 순간유량 값의 평균 오차 값을 산출하여 95% 이상 정확도를 갖는지를 시험할 수 있다.In transmitting the IoT data in the above-described step S401, in the case of measuring the flow rate data in the IoT complex valve, in the data collector 110, after executing 100% of the opening degree through the control device connected to the IoT complex valve, the medium flow rate Standard instantaneous flow rate value in test device 39

(L/min), check whether the corresponding flow rate value is monitored through the display of the control device in the IoT complex valve, and after repeating this three times, the average error value of the instantaneous flow rate value of the standard and valve for each time You can test whether it has more than 95% accuracy by calculating .

In transmitting the IoT data in the above-described step S401, in the case of measuring the pressure data in the IoT complex valve, in the data collector 110, the condition of the test device is set to 0.8 MPa and the pressure value is monitored by the display unit connected to the valve. After checking whether or not it is 5 times, the average accuracy can be calculated.

In transmitting the IoT data in the above-described step S401, in the case of measuring the pressure data in the IoT complex valve, the data collector 110 checks whether the pressure in the IoT complex valve is monitored by the display unit of the control device, and the conditions of the test device set to 20°C, 0.8 MPa to check whether the pressure value is accurately measured by the display of the control device, set the condition of the test device to 20°C, 1.6 MPa to check if it is monitored by the display, and check this 5 times. After repeated execution, the measured pressure value can be tested for accuracy measured within ±0.5% error range of the reference pressure value.

In transmitting the IoT data in the above-described step S401, in the case of measuring the temperature data in the IoT complex valve, the data collector 110 sets a specific temperature in the chamber and maintains the valve in a stable state for 2 hours, and then displays the temperature to the display unit. After checking 5 times whether the value is monitored, the average accuracy can be calculated.

In transmitting the IoT data in the above-described step S401, in the case of measuring temperature data in the IoT complex valve, in the data collector 110, after setting the IoT complex valve and test device environment to (0.0 ± 3.0) ° C, 1 hour While maintaining the IoT complex valve in a stable state, checking whether it is monitored by the display of the control device, setting the temperature of the test device to (40±3) °C, and maintaining the IoT complex valve in a stable state for 2 hours. After checking the measured value, and after repeating this 5 times, it is possible to calculate the accuracy after confirming that the temperature measured value of the display part maintains within the error range of ±1.0°C.

In transmitting the IoT data in step S401 described above, in the case of smart DDC input measurement device data monitoring, the data collector 110 determines whether the data input values of heterogeneous measurement facilities (eg, power, air conditioning, etc.) are monitored. After checking more than 5 times, the average accuracy can be calculated.

In transmitting the IoT data in the above-described step S401, in the case of smart DDC input measurement device data monitoring, in the data collector 110, the data detected by the instrument of the building facility (eg, power, air conditioning, etc.) is transmitted through the communication network. It is transmitted to the smart DDC through smart DDC, and it is checked whether the data input value is monitored by the building equipment (electric power, air conditioning, etc.) connected to the smart DDC. The average accuracy can be calculated after saving the data using real-time storage or using a sensor-based simulation tool and checking whether the data input value is monitored more than 5 times.

In transmitting the IoT data in the above-described step S401, in the case of the IoT measurement data collection and storage algorithm, the data collector 110 selects 5 types of data to be collected and 5 types of data to be stored from the IoT sensor device, thereby increasing the accuracy of the algorithm. can be tested.

In transmitting the IoT data in step S401 described above, in the case of the IoT measurement data collection and storage algorithm, in the data collector 110, after executing the IoT measurement data collection algorithm, five types of actual data are normally received from the IoT complex valve. You can test the accuracy of the algorithm by testing whether it collects data and running 10 iteration tests.

In transmitting the IoT data in the above-described step S401, in the case of the predictive maintenance algorithm, in the data collector 110, the failure type (eg, IoT complex valve flow rate, pressure, temperature data measurement device failure, control drive failure, Smart DDC communication transmission failure, etc.) can be used to measure the accuracy error rate (CvRMSE) of the predictive maintenance algorithm using 10 types of test data.

In transmitting the IoT data in the above-described step S401, in the case of the predictive maintenance algorithm, in the data collector 110, after executing the federated learning-based predictive maintenance algorithm, the collected and stored IoT measurement data and the existing maintenance data ( For example, in the case of an IoT complex valve, whether or not the response and data update are checked according to the data transmission/reception cycle (up to 30 seconds), and in the case of the control drive unit, whether the return signal of the current state and the control signal from the facility match Check the control status, and in case of smart DDC, check whether communication is transmitted/received at regular intervals (1 second) in conjunction with the supervisory control monitoring system, etc.) to train to build a predictive model, , IoT complex valve flow rate, pressure, temperature data measurement device failure, control drive failure, smart DDC communication transmission failure, etc.) After confirming the prediction results for 10 types of test data, 30% of the total data was used to predict the prediction model. After checking the error rate (CvRMSE) more than 5 times, the average error rate can be measured.

In transmitting the IoT data in step S401 described above, in the case of the federated learning-based predictive maintenance algorithm, the data collector 110 may measure the processing speed of the predictive maintenance algorithm using 10 types of test data.

In transmitting the IoT data in step S401 described above, in the case of the federated learning-based predictive maintenance algorithm, the data collector 110 trains the collected and stored IoT measurement data after executing the federated learning-based predictive maintenance algorithm. After building the predictive model, we can measure the prediction time for 10 types of test data.

In transmitting the IoT data in step S401 described above, in the case of an integrated solution for predictive maintenance, in the data collector 110, the composite solution for predictive maintenance service uses 10 types of test data to predict without errors. You can test the stability of what you do.

In transmitting the IoT data in the above-described step S401, in the case of an integrated solution for predictive maintenance, the data collector 110 executes the predictive maintenance integrated solution using IoT measurement data, and then collects 10 types of test data. Confirm the derivation of the prediction result for , and test the stability of the software through 10 repeated tests.

In transmitting the IoT data in the above-described step S401, the data collector 110 classifies the data for mechanical equipment (automatic control) failure abnormalities, measurement signal errors, communication state errors, reverse flow errors, input sensor errors, and inoperability. etc. can be done.

In transmitting the IoT data in the above-described step S401, the data collector 110 can collect and store real-time data from the IoT device (eg, IoT complex valve), in the flow rate, temperature, and pressure sensor of the IoT complex valve. It can collect real-time flow, temperature, and pressure data, and uses Apache Flume for real-time data collection, HDFS, HBase, and Impala to manage big data in a distributed environment, and real-time data analysis and processing. You can also use Apache Spark for this.

When the IoT data is delivered in the above-described step S401, the federated learner 120 performs federated learning on the IoT data transmitted from the data collector 110 to generate federated learning data to the AI predictive analyzer 130. It will be delivered (S402).

In transmitting the federated learning data in the above-described step S402, the federated learner 120 applies a federated learning technique, which is a machine learning technique, to link to the server 121 and the server 121 rather than learning on a single server. A plurality of edge devices may learn IoT data transmitted from the data collector 110 by distributing roles to each other.

In transmitting the federated learning data in step S402 described above, the federated learner 120 can build a federated learning system including the server 121 and a plurality of DDCs 122 in order to apply the federated learning technique. , At this time, it is possible to perform federated learning on IoT data by mounting a processor capable of storing and calculating IoT data in the DDC 122.

In transmitting the joint learning data in step S402 described above, in the DDC 122, a digital value (eg, value or communication) directly from a specific location (eg, central monitoring unit, disaster prevention room, etc.) to a device in the field can control.

In transmitting the federated learning data in the above-described step S402, the DDC 122 can store and process data, and apply federated learning based on this to support fast control and calculation speed, so that the AI predictive analyzer 130 enables real-time failure prediction.

In transmitting the federated learning data in step S402 described above, the federated learner 120 can perform preprocessing analysis and control at the edge device level, using federated learning and distributed processing edge computing technology by the edge device. By applying, it is possible to learn the pre-processing analysis in the previous step of the AI predictive analyzer 130.

In transmitting the federated learning data in step S402 described above, in the federated learner 120, the IoT data transmitted from the data collector 110 may be primarily learned through preprocessing analysis.

In transmitting the federated learning data in step S402 described above, the federated learner 120 primarily preprocesses and analyzes the plurality of sensing information transmitted from the data collector 110 and compares them with other sensing values in real time to recognize temporary disability. You can check if there is a problem or not.

In transmitting the federated learning data in the above-described step S402, the DDC 122 can primarily diagnose the failure of the machine. At this time, the failure is compared with the learned data of the server 121 based on the collected data. diagnosis can be made.

In transmitting the federated learning data in step S402 described above, the federated learner 120 may classify common conditions for failure prediction analysis through federated learning to find a hardware common use analysis model.

When the federated learning data is transmitted in step S402 described above, the AI predictive analyzer 130 predictively analyzes the failure of the machine in real time based on AI using the federated learning data transmitted from the federated learner 120 ( S403).

In predicting and analyzing the failure of mechanical facilities in the above-described step S403, in the AI predictive analyzer 130, it is possible to predict and manage failures of mechanical facilities in real time by applying AI techniques to management of automatic control facilities. It can detect, analyze, control and manage the occurrence of abnormalities in mechanical facilities.

In the predictive analysis of the failure of the machine in step S403 described above, in the AI predictive analyzer 130, by applying analysis techniques such as logic, decision-making, rules, and models to the federated learning data transmitted from the federated learner 120 , can help improve productivity and efficiency.

In the above-described step S403, in the predictive analysis of the failure of the machine, the AI predictive analyzer 130 uses the federated learning data transmitted from the federated learner 120 (ie, primarily preprocessed and analyzed learning data) to achieve this. A corresponding rule (that is, the most suitable and efficient rule) can be created. At this time, pre-defined rules may be remodeled using the first pre-processed and analyzed learning data.

In predicting and analyzing the failure of machinery in the above-described step S403, the AI predictive analyzer 130 does not derive and predict an energy usage-based pattern, but models through analysis of the learning data that is primarily pre-processed and analyzed. predictive service can be provided.

In the predictive analysis of the failure of the machine in step S403 described above, the AI predictive analyzer 130 does not detect and analyze abnormal conditions using only predefined rules, but the federated learning transmitted from the federated learner 120. AI-based CBM (condition based maintenance) using data (i.e., primary pre-processing analysis learning data) to regularly check the condition of machinery and equipment to identify precursors of failure, and predict preceding failure before reaching the final failure Predictive maintenance can be performed.

In the above-described step S403, in the predictive analysis of the failure of the mechanical equipment, the AI predictive analyzer 130 provides basic combined learning data for predicting the failure of the mechanical equipment even if the failure of the mechanical equipment has never occurred. Big data can be databased in advance by failure type and stored in memory, and the failure of mechanical equipment can be predicted by comparing and analyzing the federated learning data transmitted from the federated learner 120 with the big data stored in the memory, In addition, the reliability of big data can be improved by continuously upgrading the corresponding new federated learning data in the memory whenever a machine failure occurs.

In predicting and analyzing the failure of machinery in step S403 described above, in the AI predictive analyzer 130, failure of machinery and A/S situations (eg, IoT complex valves, equipment control monitoring devices, digital instruments, etc. ) may be modeled and stored in memory.

In the predictive analysis of the failure of the hardware in step S403 described above, the AI predictive analyzer 130 may provide multiple coordinated visualization of the analyzed data to the user.

In the predictive analysis of failures of machinery in the above-described step S403, in the AI predictive analyzer 130, when providing predictive maintenance service, as shown in FIG. 3, it can be divided into daily inspection and monitoring. In the case of inspection, when auto gatering is not possible, it is possible to distinguish whether the collected data is judgment data or weighing data. In the case of weighing data, the data collected through the measuring device or sensor 111 can be used to diagnose based on the pre-set upper and lower specification limits,

In the predictive analysis of the mechanical failure in the above-described step S403, the AI predictive analyzer 130 can distinguish whether the collected data is digital data or analog data in the case where auto gating is possible during monitoring, and digital data In this case, the state signal of the measuring device or sensor 111 may be collected to diagnose whether it is 1 (ie, abnormal) or 0 (ie, normal), and in the case of analog data, the signal of the measuring device or sensor 111 The measured values may be collected and diagnosed based on a pre-set upper and lower specification limit line, or may be diagnosed based on a management limit line using a control chart pattern.

In predicting and analyzing the failure of the mechanical equipment in the above-described step S403, in the AI predictive analyzer 130, the type of failure of the automatic control equipment (eg, measurement due to the failure of the IoT complex valve data measurement device (flow rate, pressure, temperature) In order to perform predictive analysis according to data update transmission failure, flow and temperature control failure due to failure of IoT complex valve control drive unit, failure of facility data monitoring and control due to smart DDC communication transmission/reception failure, etc.), failure of IoT composite valve data measurement device ( flow rate, pressure, temperature), you can check the response and data update according to the data transmission/reception cycle (up to 30 seconds), and in case of failure of the IoT complex valve control drive unit, It is possible to check the control status through matching, and in case of a smart DDC communication transmission/reception failure, it is possible to check whether communication is transmitted/received at regular intervals (eg, 1 second) in connection with the monitoring control monitoring system.

In predicting and analyzing the failure of mechanical equipment in the above-described step S403, in the AI predictive analyzer 130, in order to predict and determine the failure situation of the heating and cooling system of the building, the measured values of the components are different from the average measurement and If an error of more than a certain level (eg, 5%) is shown, maintenance is required, if emergency information is collected above a certain level (eg, about 30% of components), it can be classified as a need for full-scale system repair.

As described above, the embodiments of the present invention are not implemented only through the above-described device and/or operating method, but through a program for realizing functions corresponding to the configuration of the embodiment of the present invention and a recording medium on which the program is recorded. It may be implemented, and such an implementation can be easily implemented by an expert in the technical field to which the present invention belongs based on the description of the above-described embodiment. Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concept of the present invention defined in the following claims are also included in the scope of the present invention. that fall within the scope of the right.

100: Hardware failure prediction analysis system using federated learning
110: data collector
111: measuring device or sensor
120: federated learner
121: server
122: DDC
130: AI predictive analyzer

Claims (5)

a data collector for collecting IoT data from machinery;
A federated learner for receiving the IoT data collected by the data collector and performing federated learning to generate federated learning data; and
A hardware failure prediction analysis system using federated learning including an AI predictive analyzer for real-time predictive analysis of machine failure based on AI by receiving federated learning data generated by the federated learner.
The method of claim 1, wherein the data collector,
A mechanical equipment failure prediction analysis system using federated learning, characterized in that it collects IoT data measured or detected by a measuring device or sensor installed in an automatic control facility.
The method of claim 1, wherein the data collector,
In the case of flow data measurement in the IoT complex valve, a 100% opening command is executed through the control device, and after confirming three times whether the standard instantaneous flow rate value in the flow test device is monitored, using federated learning, characterized in that the average accuracy is calculated. Hardware failure prediction analysis system.
The method of claim 1, wherein the federated learner,
By applying a federated learning technique, a server and a plurality of edge devices connected to the server distribute roles to each other to receive and learn the IoT data collected by the data collector. Machinery failure prediction analysis system using federated learning, characterized in that .
Collecting, by the data collector, IoT data from the hardware;
A step in which the federated learner receives the IoT data collected by the data collector and performs federated learning to generate federated learning data; and
A method for predicting and analyzing hardware failures using federated learning, comprising the step of allowing the AI predictive analyzer to receive federated learning data generated by the federated learner and perform predictive analysis of machine failures in real time based on AI.

Images