CN111783856A - A manufacturing-oriented equipment fault auxiliary diagnosis method and system - Google Patents

Abstract

The invention discloses a manufacturing-oriented equipment fault auxiliary diagnosis method and system. The method includes but is not limited to the following processes: generating a predicted value of the operation state of the equipment to be diagnosed; monitoring the real value of the operation state of the equipment to be diagnosed in real time; The difference between the predicted value and the actual value determines whether the equipment to be diagnosed is faulty. The system includes, but is not limited to, a state prediction module, a state monitoring module, and a fault judgment module; the state prediction module is used to generate a predicted value of the operating state of the equipment to be diagnosed, and the state monitoring module is used to monitor the real value of the operating state of the equipment to be diagnosed in real time. The judgment module is used for judging whether the equipment to be diagnosed is faulty based on the difference between the predicted value and the actual value. The invention can effectively assist maintenance personnel to judge whether the monitored equipment is faulty, and has the advantages of short fault diagnosis time, low labor cost, little dependence on personnel experience and technical level, etc., thereby improving the equipment operation and maintenance efficiency and production efficiency of the manufacturing industry. .

Description

A manufacturing-oriented equipment fault auxiliary diagnosis method and system

technical field

The present invention relates to the technical field of machine learning, and more particularly, the present invention relates to a manufacturing-oriented equipment fault auxiliary diagnosis method and system.

Background technique

At present, more and more manufacturing enterprises have begun to implement digital, automated and intelligent transformation, thereby improving manufacturing efficiency and quality, and reducing operating costs and labor costs. It can be seen that a high-stability and high-reliability manufacturing system will be an important guarantee for enterprises to improve production efficiency.

The equipment fault diagnosis of manufacturing enterprises often relies on manual inspection and maintenance. The inspection personnel need to inspect the machine regularly every day, and report the fault to the maintenance personnel. The maintenance personnel take different measures according to the size of the fault, and then feedback to the workshop director, and then report to Factory manager, etc. Obviously, the whole process of manual inspection method consumes a lot of human resources, and the labor cost is high, and it has high requirements and dependence on the experience and technical level of the inspectors, and the equipment diagnosis time is long, the maintenance cycle is short, The maintenance efficiency is low, and the operation status of the equipment cannot be tracked in real time. Therefore, the existing manual inspection and maintenance scheme cannot guarantee the stability and reliability of the manufacturing system, and cannot realize the digitalization, automation and intelligent transformation of manufacturing enterprises.

Although some people have proposed an intelligent diagnosis scheme for equipment faults, due to its technical limitations, there are still problems such as high labor costs, greater dependence on personnel experience and technical level, and long equipment fault diagnosis time.

SUMMARY OF THE INVENTION

In order to solve the problems of high labor cost, long fault diagnosis time, and excessive dependence on personnel in the existing fault diagnosis scheme of manufacturing equipment, the present invention innovatively provides a manufacturing equipment fault auxiliary diagnosis method and system. .

In order to achieve the above technical purpose, the present invention specifically discloses a manufacturing-oriented equipment fault auxiliary diagnosis method, which includes but is not limited to the following processes: generating a predicted value of the operation state of the equipment to be diagnosed; monitoring the operation of the equipment to be diagnosed in real time The actual value of the state; based on the difference between the predicted value and the actual value, determine whether the device to be diagnosed is faulty.

Further, the process of generating the predicted value of the operating state of the equipment to be diagnosed includes but is not limited to the following processes: collecting historical real-time operating data of the equipment to be diagnosed at a set frequency; distribution features, and generate time series features of historical real-time operation data within a second preset time period; use the distribution features to train a feature classifier, so as to predict the first feature value of future operating conditions through the feature classifier; use the The time series feature trains a feature regressor to predict a second feature value of future operating conditions through the feature regressor; the first feature value and the second feature value are feature-fused to generate the predicted value.

Further, the process of judging whether the device to be diagnosed is faulty includes but is not limited to the following process: taking the difference between the predicted value and the actual value as the input of the logistic regression classifier, and judging the fault based on the output of the logistic regression classifier. Describe whether the equipment to be diagnosed is faulty; the training set of the logistic regression classifier includes training data with labeled samples.

Further, the method further includes: when judging the fault of the equipment to be diagnosed, calculating the feature weights of each real value corresponding to the current fault, and then providing the user with visual inspection reference data according to the feature weight and each real value.

Further, the time series feature is generated in a time series sliding window manner, and the duration of the sliding window is less than or equal to the second preset duration.

Further, the distribution feature is a low frequency feature, and the time sequence feature is a high frequency feature; and the first preset duration is greater than the second preset duration.

Further, the logistic regression classifier is trained as follows:

S1, use the logistic regression classifier to classify the samples of the current active learning sample set, so as to obtain the classification results of the samples in each active learning sample set;

S2, manually label the samples whose classification result probability is in the preset interval, and add the labeled samples to the training set;

S3, use all the currently labeled training samples to adjust the parameters of the current logistic regression classifier to update the current logistic regression classifier;

S4 , verify whether the classification accuracy of the current logistic regression classifier reaches the set value or there are no unlabeled samples in the active learning sample set, then the iteration is ended, otherwise steps S1 to S4 are executed in a loop.

In order to achieve the above technical purpose, the present invention also discloses a manufacturing-oriented equipment fault auxiliary diagnosis system, which includes but is not limited to a state prediction module, a state monitoring module and a fault judgment module. The state prediction module is used to generate the predicted value of the operation state of the equipment to be diagnosed; the state monitoring module is used to monitor the real value of the operation state of the equipment to be diagnosed in real time; the fault judgment module is used to based on the predicted value and the real value The difference of the values determines whether the device to be diagnosed is faulty.

Further, the state prediction module includes but is not limited to a data acquisition unit, a feature extraction unit, a preliminary prediction unit and a feature fusion unit. The data collection unit is used to collect historical real-time operation data of the equipment to be diagnosed at a set frequency; the feature extraction unit is used to count the distribution characteristics of the historical real-time operation data within the first preset time period, and is used to generate the second preset time period. time series features of historical real-time operation data within a time period; a preliminary prediction unit for training a feature classifier by using the distribution features, so as to predict the first feature value of future operating conditions through the feature classifier; and for using the time series The feature trains a feature regressor to predict the second feature value of the future operation through the feature regressor; a feature fusion unit is configured to perform feature fusion on the first feature value and the second feature value to generate the the predicted value.

Further, the fault judging module is used to use the difference between the predicted value and the actual value as the input of the logistic regression classifier, and judge whether the equipment to be diagnosed is faulty according to the output of the logistic regression classifier; logistic regression The training set of the classifier consists of training data with labeled samples.

Further, the system also includes a checking reference module. The troubleshooting reference module is configured to calculate the feature weights of each real value corresponding to the current fault when the equipment to be diagnosed is faulty, and to provide a user with visual troubleshooting reference data according to the feature weights and the respective real values.

The beneficial effects of the present invention are as follows: the present invention can effectively assist maintenance personnel to judge whether the monitored equipment is faulty, and has the advantages of short fault diagnosis time, low labor cost, and little dependence on personnel experience and technical level, and thus can improve the manufacturing industry. equipment operation and maintenance efficiency and production efficiency.

Aiming at the monitoring of manufacturing related equipment, the present invention can monitor the running status of the equipment in real time, diagnose the cause of equipment failure efficiently, and present the running deviation of the equipment in real time, so as to ensure the normal operation of the manufacturing system. Moreover, the equipment state information collected by the present invention has continuity and has high data mining value.

In addition, in view of the problem that it is difficult for manufacturing enterprises to manually organize and label equipment abnormal data by acquiring a large amount of real abnormal data, the present invention proposes an abnormal detection scheme based on active learning, which significantly reduces the labor cost of traditional manual labeling.

Description of drawings

FIG. 1 shows a schematic flowchart of a manufacturing-oriented equipment fault assistant diagnosis method.

FIG. 2 shows a schematic diagram of the working principle of a manufacturing-oriented equipment fault auxiliary diagnosis system according to some embodiments of the present invention.

Detailed ways

A manufacturing-oriented equipment fault auxiliary diagnosis method and system provided by the present invention will be explained and described in detail below with reference to the accompanying drawings.

Example 1:

As shown in FIGS. 1 and 2 , in order to solve at least one problem existing in the prior art, the present embodiment can provide a manufacturing-oriented equipment fault auxiliary diagnosis method, which can realize real-time monitoring of equipment operating conditions and efficient assistance of equipment faults. diagnosis. The method includes, but is not limited to, the following processes.

First, the historical real-time operation data of the equipment to be diagnosed is collected at a set frequency. The set frequency can be set according to actual needs, such as 0.1 Hz, etc. In this embodiment, real-time data of different sensors of the device can be acquired every 10s according to the difference of the device. For example, the corresponding real-time operation data of the machine can be collected through various sensors. The real-time operation data includes but not limited to pressure, temperature, rotational speed, etc. During implementation, pressure data can be collected through a pressure sensor, and temperature can be collected through a temperature sensor.

Secondly, the present invention obtains corresponding time sequence characteristics and distribution characteristics according to the collected real-time operation data of the equipment, so as to analyze and process the equipment operation data by frequency division. The multi-frequency time series feature processing process can be performed synchronously, and the specific description is as follows.

Count the distribution characteristics of historical real-time operation data within the first preset time period, for example, analyze the distribution range of different sensor data values of different devices according to daily frequency statistics, and then distribute the device values into different feature buckets according to the range, which is a low-frequency feature. Predictions do data preparation. That is, the distribution feature in this embodiment is a low-frequency feature, and the first preset duration may be, for example, one day. An example is as follows: For a device on a small part production line, its rotational speed is the number of parts produced divided by the production time, and its rotational speed range is roughly [0 pieces/minute, 60 pieces/minute]. Divided into three characteristic buckets: less than 10/min, between 10/min and 30/min, and more than 30/min.

A time series feature of historical real-time operation data within a second preset time period is generated, and the first preset time period is often greater than the second preset time period. The time series feature is generated in a time series sliding window manner, and the duration of the sliding window may be less than or equal to the second preset duration. In this example, a sliding window is used to take 10 minutes as the duration of the sliding window to cut the data into a sequence of time windows. It is preferable to take the average feature value of each minute as the feature value of the minute, and finally obtain the features of 10 sequences within 10 minutes. _The value [w ₁ , w ₂ , .

和第二特征值

Third, by training the machine learning model to predict the real-time operation data of the equipment, and at the same time monitor the operation of the equipment in real time. The process of predicting equipment operation data may include a basic learner learning stage and a meta-learner learning stage, and the first feature value is obtained through the basic learner learning

and the second eigenvalue

本实施例的特征分类器优选为低频的贝叶斯特征分类器。贝叶斯特征分类器是从日频统计设备在一天的不同小时的设备最可能的特征值，比如转速值。模型输入为小时值，比如上午10点，输入值即为10，最后通过分类器可预测出10点该设备的最可能的特征值：Basic learner learning stage: use the distribution features to train the feature classifier to predict the first feature value of the future operation through the feature classifier

The feature classifier in this embodiment is preferably a low-frequency Bayesian feature classifier. The Bayesian feature classifier is derived from the daily frequency statistics of the most likely feature values of the device at different hours of the day, such as rotational speed values. The model input is the hour value, such as 10 am, the input value is 10, and finally the classifier can predict the most likely feature value of the device at 10 o'clock:

用于表示预测的特征结果；x_i表示与预测特征相关的特征(即所有除当前特征的以外的其他特征)；n表示其他特征数量；P(y)表示事件y发生的概率，P(x_i|y)表示事件y发生条件下x_i的概率。in,

Used to represent the predicted feature results; _xi represents the features related to the predicted feature (that is, all other features except the current feature); n represents the number of other features; P(y) represents the probability of event y, P(x _i |y) represents the probability of x _i under the condition that event y occurs.

本实施例的特征回归器优选为高频的LightGBM(Light Gradient BoostingMachine，基于决策树算法的分布式梯度提升框架)特征回归器。本实施例的LightGBM特征回归器可从过去10分钟的特征值预测设备未来10s的特征值，比如转速值。模型输入为过去十分钟里每分钟的特征均值[w₁,w₂,…,w₁₀]，预测未来10s的特征值。LightGBM是微软2017开发的一个开源机器学习项目，高效地实现了梯度提升决策树(GBDT，GradientBoostingDecisionTree)算法，对传统的GBDT和Xgboost(eXtreme GradientBoosting，极端梯度提升)做了工程上的优化，可在不损失精度的情况下提升算法运算速度，比如采用下述的方式进行运算。Next, use the time series feature to train the feature regressor, and predict the second feature value of the future operation through the feature regressor

The feature regressor in this embodiment is preferably a high-frequency LightGBM (Light Gradient Boosting Machine, a distributed gradient boosting framework based on a decision tree algorithm) feature regressor. The LightGBM feature regressor in this embodiment can predict the feature value of the device in the next 10s, such as the rotational speed value, from the feature value of the past 10 minutes. The model input is the feature mean value [w ₁ ,w ₂ ,...,w ₁₀ ] every minute in the past ten minutes, and predicts the feature value for the next 10s. LightGBM is an open source machine learning project developed by Microsoft in 2017. It efficiently implements the gradient boosting decision tree (GBDT, GradientBoostingDecisionTree) algorithm, and has made engineering optimizations for traditional GBDT and Xgboost (eXtreme GradientBoosting, extreme gradient boosting), which can be found in To improve the operation speed of the algorithm without losing precision, for example, use the following method to perform the operation.

用于表示特征预测结果，F_i(x_j)表示第i颗决策树的预测结果，L_i表示第i颗决策树基于第i-1颗决策树训练目标。in,

It is used to represent the feature prediction result, F _i (x _j ) represents the prediction result of the ith decision tree, and _Li represents the training target of the ith decision tree based on the i-1th decision tree.

Then, the meta-learner learning stage is performed: that is, the prediction result of the fusion base learner, for example, may include but not limited to the feature value after 10s of prediction. The first eigenvalue and the second eigenvalue are feature-fused to generate a predicted value.

预测结果为10s后设备的特征值，本实施例采用ElasticNet(弹性网络)回归算法，能够将L₁正则项和L₂正则项完美结合在一起，抑制模型过拟合，提高模型泛化能力：The input of the meta-learner is the prediction result of the base learner

The prediction result is the eigenvalue of the device after 10s. The ElasticNet (elastic network) regression algorithm is used in this embodiment, which can perfectly combine the L ₁ regular term and the L ₂ regular term, suppressing the overfitting of the model and improving the generalization ability of the model:

是L₁正则项系数平衡L₁和L₂正则项，α是正则项整体的系数。where L is the loss function of the regression algorithm, W is the coefficient of the regression equation,

is the L ₁ canonical term coefficient to balance the L ₁ and L ₂ canonical terms, and α is the coefficient of the entire canonical term.

Therefore, the present invention can generate the predicted value of the operating state of the equipment to be diagnosed through the learning stage of the basic learner and the learning stage of the meta-learner, and the present invention can monitor the real value of the operating state of the equipment to be diagnosed in real time.

Finally, it is judged whether the equipment to be diagnosed is faulty based on the difference between the obtained predicted value and the actual value. Both the predicted value and the actual value involved in this embodiment can be the characteristic value of the corresponding parameter. The classification algorithm can be used to analyze whether there is an abnormality in the equipment, and the abnormal cause of the auxiliary diagnosis can be provided. More specifically, the process of judging whether the equipment to be diagnosed is faulty includes: taking the difference between the predicted value and the actual value as the input of the logistic regression classifier (equipment anomaly classifier), and judging whether the equipment to be diagnosed is not based on the output of the logistic regression classifier. Fault. As shown in FIG. 2 , this embodiment can provide anomaly detection based on active learning, including an active learning part and an auxiliary diagnosis part.

Active learning part: The goal is to learn a good enough classification module with as few labeled samples as possible. The input of the equipment abnormality classifier is the difference between the feature prediction value of the multi-level feature prediction module and the real feature value, and the output is the classification result, indicating whether the current machine is abnormal.

The algorithm flow of the equipment anomaly classifier is as follows: (1) Select the logistic regression classifier and the corresponding training set (initially zero), validation set (with labeled samples), and active learning sample set (unlabeled samples); (2) Initialization Logistic regression classifier (device anomaly classifier in Figure 2) parameters; (3) Use the logistic regression classifier to classify the samples of the current active learning sample set one by one to obtain the classification results of the samples in each active learning sample set , select a sample whose classification result probability is in the preset interval, the preset interval in this embodiment is [0.4, 0.6]; the closer the classification result is to 0.5, the higher the uncertainty of the current model for the sample, and the corresponding (4) The maintenance personnel can manually label the selected samples in the preset interval, and then add the labeled samples to the training set, then the training set of the logistic regression classifier includes the training of the labeled samples. This embodiment reduces a lot of manual labeling workload than simply judging faults by labeling samples, so the present invention can effectively reduce the cost of labeling abnormal equipment data for manufacturing enterprises; (5) Use all currently labeled training samples to classify the current logistic regression (6) Verify whether the classification accuracy of the current logistic regression classifier reaches the set value, for example, use the current logistic regression classifier to perform mirror verification on the validation set , if the performance of the current logistic regression classifier reaches the classification accuracy of 80% or there are no unlabeled samples in the active learning sample set, the iteration ends, otherwise (3)-(6) are executed in a loop. The purpose of the above-mentioned active learning process is to learn a classification module on the logistic regression classifier under the premise of a small number of labeled samples as much as possible, so as to reduce the labor cost of manual labeling (labeling abnormal device data).

Auxiliary diagnosis part: The goal is to monitor the operation status of the equipment and provide equipment abnormal information to the equipment with abnormal conditions. For the real-time status data of the equipment, this embodiment uses the multi-level feature prediction model to obtain the deviation between the predicted value and the current real-time data, and visualizes each feature deviation value and provides it to the maintenance personnel to monitor the operation of the equipment; then the current iteration of the active learning module can be used. The learned model analyzes whether the current equipment is faulty. If there is a fault, the deviation of different characteristics is provided to the maintenance personnel to help diagnose which characteristics of the equipment are faulty, and then the coefficients of different characteristics of the logistic regression model are provided to help the maintenance personnel diagnose which ones are The characteristics are the most relevant to the fault and the auxiliary maintenance personnel investigate the cause. In some embodiments of the present invention, when judging the fault of the equipment to be diagnosed, the feature weight of each real value corresponding to the current fault is calculated, and then visual inspection reference data can be provided to the user according to the feature weight and each real value.

Embodiment 2:

As shown in FIG. 2 , based on the same inventive concept as the first embodiment, the present embodiment provides an auxiliary diagnosis system for equipment faults oriented to the manufacturing industry. The entire equipment fault auxiliary diagnosis system framework can include three parts: data acquisition part, multi-level feature prediction model part and anomaly detection part based on active learning, multi-level feature prediction model part can include two-level prediction, the idea comes from integrated learning stacking (Stacked) framework, divided into a base learning part and a meta learning part. Specifically, the auxiliary fault diagnosis system may include, but is not limited to, a state prediction module, a state monitoring module, a fault judgment module, and a troubleshooting reference module.

The state prediction module is used to generate the predicted value of the operating state of the equipment to be diagnosed. The state prediction module includes but is not limited to a data acquisition unit, a feature extraction unit, a preliminary prediction unit and a feature fusion unit.

The data collection unit is used to collect historical real-time operation data of the equipment to be diagnosed at a set frequency.

The feature extraction unit is used to count the distribution characteristics of the historical real-time operation data within the first preset time period, and is used to generate the time series characteristics of the historical real-time operation data within the second preset period of time; in this embodiment, a time-series sliding window method can be used A time series feature is generated, and the duration of the sliding window is less than or equal to the second preset duration. The distribution feature is a low frequency feature, the time sequence feature is a high frequency feature, and the first preset duration is greater than the second preset duration.

The preliminary prediction unit is used for training the feature classifier by using the distribution feature, so as to predict the first feature value of the future operation condition by the feature classifier; second eigenvalue.

The feature fusion unit is used for feature fusion of the first feature value and the second feature value to generate a predicted value.

The status monitoring module is used to monitor the real value of the running status of the equipment to be diagnosed in real time.

The fault judgment module is used for judging whether the equipment to be diagnosed is faulty based on the difference between the predicted value and the actual value. The fault judgment module is also used to use the difference between the predicted value and the actual value as the input of the logistic regression classifier, and judge whether the equipment to be diagnosed is faulty according to the output of the logistic regression classifier; the training set of the logistic regression classifier includes the training with labeled samples. data.

The troubleshooting reference module is used to calculate the feature weight of each real value corresponding to the current fault when the equipment to be diagnosed is faulty, and to provide the user with visual troubleshooting reference data according to the feature weight and each real value.

Logic and/or steps represented in flowcharts or otherwise described herein, for example, may be considered an ordered listing of executable instructions for implementing the logical functions, and may be embodied in any computer-readable storage medium , for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, device or equipment. For the purposes of this specification, a "computer-readable storage medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or apparatus . More specific examples (non-exhaustive list) of computer readable storage media include the following: electrical connections with one or more wiring (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM, Random Access Memory), Read-Only Memory (ROM, Read-Only Memory), Erasable Programmable Read-Only Memory (EPROM, Erasable Programmable Read-Only Memory, or Flash Memory), Optical Devices, and Portable Optical Disc Read-Only Memory (CDROM, Compact Disc Read-Only Memory). In addition, the computer-readable storage medium may even be paper or other suitable medium on which the program can be printed, as the paper or other medium may be optically scanned, for example, and then edited, interpreted or, if necessary, otherwise Process in a suitable manner to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present invention may be implemented in hardware, software, firmware or a combination thereof. In the above-described embodiments, various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or a combination of the following techniques known in the art: Discrete logic circuits, application-specific integrated circuits with suitable combinational logic gate circuits, Programmable Gate Array (PGA, Programmable Gate Array), Field Programmable Gate Array (FPGA, Field Programmable Gate Array), etc.

In the description of this specification, reference to the terms "this embodiment", "one embodiment", "some embodiments", "example", "specific example", or "some examples" or the like is meant to be combined with the description of the embodiment A particular feature, structure, material or characteristic described or exemplified is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, those skilled in the art may combine and combine the different embodiments or examples described in this specification, as well as the features of the different embodiments or examples, without conflicting each other.

In addition, the terms "first" and "second" are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature delimited with "first", "second" may expressly or implicitly include at least one of that feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise expressly and specifically defined.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and simple improvements made in the essence of the present invention should be included in the protection scope of the present invention. Inside.

Claims (10)

1. A manufacturing-oriented equipment fault auxiliary diagnosis method is characterized by comprising the following steps:

generating a predicted value of the running state of the equipment to be diagnosed;

monitoring the real value of the running state of the equipment to be diagnosed in real time;

and judging whether the equipment to be diagnosed is in fault or not based on the difference value between the predicted value and the true value.

2. The method for assisting in diagnosing the equipment failure in the manufacturing industry according to claim 1, wherein the step of generating the predicted value of the operation state of the equipment to be diagnosed includes:

collecting historical real-time operation data of equipment to be diagnosed at a set frequency;

counting the distribution characteristics of the historical real-time operation data within a first preset time length, and generating the time sequence characteristics of the historical real-time operation data within a second preset time length;

training a feature classifier by using the distribution features to predict a first feature value of a future operating condition through the feature classifier; training a feature regressor by using the time sequence features to predict a second feature value of a future operating condition through the feature regressor;

and performing feature fusion on the first feature value and the second feature value to generate the predicted value.

3. The auxiliary diagnosis method for equipment failure in manufacturing industry according to claim 1 or 2, wherein the process of determining whether the equipment to be diagnosed is failed comprises:

and taking the difference value between the predicted value and the true value as the input of a logistic regression classifier, and judging whether the equipment to be diagnosed is in fault or not according to the output of the logistic regression classifier.

4. The auxiliary diagnosis method for equipment failure in manufacturing industry according to claim 1 or 2, further comprising:

and when the equipment to be diagnosed is judged to be in fault, calculating the characteristic weight of each real value corresponding to the current fault, and then providing visual investigation reference data for a user according to the characteristic weight and each real value.

5. The auxiliary diagnosis method for equipment fault facing manufacturing industry of claim 2, wherein the time sequence feature is generated by adopting a time sequence sliding window mode, and the duration of the sliding window is less than or equal to the second preset duration.

6. The auxiliary diagnosis method for equipment failure in manufacturing industry according to claim 3,

training the logistic regression classifier by:

s1, classifying the samples of the current active learning sample set by using a logistic regression classifier to obtain the classification result of the samples in each active learning sample set;

s2, manually marking the samples with the classification result probability in a preset interval, and adding the marked samples into a training set;

s3, using all the marked training samples to adjust the parameters of the current logistic regression classifier so as to update the current logistic regression classifier;

and S4, verifying whether the classification accuracy of the current logistic regression classifier reaches a set value or whether no unlabeled sample exists in the active learning sample set, ending iteration, and otherwise, executing S1-S4 in a circulating mode.

7. A manufacturing-oriented equipment failure auxiliary diagnostic system, comprising:

the state prediction module is used for generating a predicted value of the running state of the equipment to be diagnosed;

the state monitoring module is used for monitoring the real value of the running state of the equipment to be diagnosed in real time;

and the fault judgment module is used for judging whether the equipment to be diagnosed has faults or not based on the difference value between the predicted value and the true value.

8. The manufacturing-oriented equipment fault assisted diagnosis system of claim 7, wherein the state prediction module comprises:

the data acquisition unit is used for acquiring historical real-time operation data of the equipment to be diagnosed at a set frequency;

the characteristic extraction unit is used for counting the distribution characteristics of the historical real-time operation data in a first preset time length and generating the time sequence characteristics of the historical real-time operation data in a second preset time length;

a preliminary prediction unit, configured to train a feature classifier using the distribution features to predict a first feature value of a future operating condition by the feature classifier; and a second feature value for training a feature regressor using the timing features to predict future operating conditions by the feature regressor;

and the feature fusion unit is used for performing feature fusion on the first feature value and the second feature value to generate the predicted value.

9. The manufacturing-oriented equipment failure auxiliary diagnostic system of claim 7 or 8,

the fault judgment module is used for taking the difference value between the predicted value and the true value as the input of a logistic regression classifier and judging whether the equipment to be diagnosed is in fault or not according to the output of the logistic regression classifier; the training set of the logistic regression classifier includes labeled sample training data.

10. The manufacturing-oriented equipment fault assisted diagnostic system of claim 7, further comprising:

and the troubleshooting reference module is used for calculating the characteristic weight of each real value corresponding to the current fault when the equipment to be diagnosed fails, and providing visual troubleshooting reference data for a user according to the characteristic weight and each real value.

Images