Feature extraction optimization method, fault prediction method, processor and storage medium
Through the multi-feature extraction model combination and iterative optimization technology, the problems of poor universality and insufficient generalization performance of feature extraction methods in the existing technology are solved, effective extraction and fault diagnosis of multi-working data are realized, and fault detection is improved Efficiency and accuracy.
Patent Information
- Application Number
- CN202411180026.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-27
AI Technical Summary
The existing feature extraction methods are not universal and have poor generalization performance, and cannot effectively extract multi-condition data. They lack iterable and self-optimized characteristics, making it difficult to adapt to fault diagnosis of different operating conditions.
Using multi-feature extraction model combination, by preprocessing data and matching feature extraction models, obtaining multi-dimensional feature sets, combining historical data for clustering and classifier training, adjusting model parameters in real time, and using fuzzy control algorithm to optimize weight ratio to realize feature extraction Iterative optimization of the model.
It improves the universality and generalization performance of feature extraction, can adapt to multi-condition data, enhances the monitoring and fault detection efficiency of real-time data, and provides targeted fault diagnosis capabilities.
Smart Images

Figure CN118690179B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of fault diagnosis, and in particular to a feature extraction optimization method, a fault prediction method, a processor and a storage medium. Background Art
[0002] With the rapid development of industrial modernization, the functions and structures of mechanical equipment are becoming more and more complex. In actual production, if mechanical equipment fails and is not repaired in time, it may cause an accident and paralyze the entire industrial chain. Therefore, timely diagnosis of mechanical equipment is necessary and urgent.
[0003] Fault diagnosis relies on information collected from sensors, such as vibration, temperature, oil debris, etc., so it is extremely important to extract valuable feature information from the original signal. Feature engineering is to convert the collected raw data into feature variables that can better express the essence of the problem, which is an important part of fault diagnosis. Feature engineering greatly affects the prediction performance of the classifier and plays a vital role in the results of fault diagnosis.
[0004] The existing feature extraction methods mainly include extraction methods based on time domain, frequency domain, time-frequency domain, etc., including short-time Fourier transform, continuous wavelet transform, etc. This type of feature extraction method is not universal, and the feature extraction system has poor generalization performance. At the same time, only simple fault identification can be performed on the real-time collected data, and it is impossible to reuse the real-time collected data to optimize the feature extraction model. In addition, the existing feature extraction method does not have the characteristics of iterative and self-optimization features. It is mainly aimed at a single working condition. When the operating parameters of different working conditions are different and the data feature distribution is inconsistent, it is difficult to extract targeted features to cope with multiple working conditions. The extracted features are relatively single and have poor practicality.
[0005] In view of this, there is an urgent need for a feature extraction optimization method that can solve the above problems. Summary of the invention
[0006] In order to solve the above technical problems, the purpose of the embodiments of the present application is to provide a feature extraction optimization method, a fault prediction method, a processor and a storage medium.
[0007] In order to achieve the above objectives, the present application provides a first aspect of a feature extraction optimization method, comprising:
[0008] Step 10, obtaining the pre-processed equipment operating condition data;
[0009] Step 20, matching at least one feature extraction model according to the equipment operating condition data;
[0010] Step 30, obtaining a multidimensional feature set of equipment operating condition data outputted by at least one feature extraction model;
[0011] Step 40, inputting the multidimensional feature set and historical equipment operating condition data into the trained classifier model;
[0012] Step 50, obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model;
[0013] Step 60, adjusting parameters of the feature extraction model according to the feature ranking;
[0014] Evaluate whether the performance of the feature extraction model after adjusting the parameters meets the preset requirements. When the performance of the feature extraction model does not meet the preset requirements, repeat steps 10 to 60.
[0015] Preferably, the method further comprises:
[0016] Get historical data sets of the device , the historical data set The data in the table has been marked with equipment operation data;
[0017] The historical data set is used The mean algorithm calculates cluster centers;
[0018] For the historical data set Clustering ,in is a fault-free data cluster, is the fault data cluster;
[0019] Set the cluster safety area, with the cluster center as the dot and the circle with the distance of the data point farthest from the cluster center as the radius as the safety area.
[0020] Preferably, before the step of obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model, the method further includes:
[0021] Performing clustering processing on the multidimensional feature set;
[0022] Acquire the operation data of the device in real time, where the operation data is unlabeled data, calculate the distance between each unlabeled data and the center of each cluster, and assign the unlabeled data to the nearest cluster;
[0023] Set data points outside the cluster safety area as warning points;
[0024] Use the trained classifier to make predictions for data outside the cluster safety zone;
[0025] If the predicted classification result is normal data, updating the cluster safety area;
[0026] If the classification result is faulty data, the faulty data is removed;
[0027] The updated cluster centers are recalculated using the data of all labeled equipment operation data and unlabeled data.
[0028] Preferably, the step of evaluating whether the performance of the feature extraction model after adjusting the parameters meets the preset requirements comprises:
[0029] The comparison loss function is calculated based on the fault diagnosis prediction result output by the classifier model and the clustering result, where the loss function is:
[0030] ;in is the temperature parameter, which is used to adjust the uniformity of the probability distribution. is the batch size, represents the cosine similarity, , , Represents the output prediction two-dimensional vector of the model. Vector 1 is the probability of predicted failure, and vector 2 is the probability of predicted non-failure. The clustering output is converted into vector form using one-hot encoding.
[0031] Preferably, the step of obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model includes:
[0032] Training the multidimensional feature set according to the classifier model, and obtaining an evaluation result of the classifier model evaluating the features of the multidimensional feature set according to the evaluation index;
[0033] Obtaining importance rankings of all features of the multidimensional feature set;
[0034] Eliminate the lowest ranked feature among all features.
[0035] Preferably, the step of adjusting parameters of the feature extraction model according to the feature ranking comprises:
[0036] The weight ratio of the feature extraction model is adjusted according to the fuzzy control algorithm, and the weight ratio of the feature extraction model after the parameter adjustment is updated.
[0037] Preferably, the step of adjusting the parameters of the weight proportion of the feature extraction model according to the fuzzy control algorithm includes:
[0038] Define the input as feature ranking and the output as model weight ratio;
[0039] Fuzzify the ranking and define three fuzzy sets: high ranking: top 30%; middle ranking: between 30% and 70%; low ranking: bottom 30%;
[0040] Define fuzzy rules: First rule: if the ranking is high, increase the weight of the corresponding feature extraction model; second rule: if the ranking is medium, adjust the weight to remain unchanged; third rule: if the ranking is low, reduce the weight of the corresponding feature extraction model;
[0041] Fuzzy reasoning, calculates the membership degree corresponding to each feature ranking according to the input feature ranking, and calculates the rule activation degree and synthesizes it;
[0042] Defuzzification, define three fuzzy sets with medium, low and high weights, and their corresponding triangular membership functions are as follows: Low weight: Triangular MF, parameters are ; Medium weight: Triangular MF, parameters are ; High weight: Triangular MF, parameters are ;in ,in is the inverse of the sort, is the hyperparameter of the weight, determined by cross-validation, middle The lowest ranking in the high ranking set is the inverse of the sort. middle The inverse of the ranking of the middle in the high ranking set, middle The highest ranking in the high ranking set is the inverse of the ranking according to the fuzzy output value; by analogy, it is calculated , ; Calculate the weight adjustment rate according to the membership function.
[0043] A second aspect of the present application provides a fault prediction method, the method comprising:
[0044] Obtain equipment operating data;
[0045] Obtain a fault prediction model by using the equipment operating condition data input value according to any of the feature extraction optimization methods described above;
[0046] The equipment fault type is outputted through the fault prediction model.
[0047] A second aspect of the present application provides a fault prediction device, comprising:
[0048] a memory configured to store instructions; and
[0049] The processor is configured to call instructions from a memory and implement the above fault prediction method when executing the instructions.
[0050] A third aspect of the present application provides a storage medium having instructions stored thereon, the instructions being used to enable a machine to execute the above-mentioned fault prediction method.
[0051] Through the above technical solution, the pre-processed equipment condition data is obtained; at least one feature extraction model is matched according to the equipment condition data; a multi-dimensional feature set of the equipment condition data outputted by at least one feature extraction model is obtained; the multi-dimensional feature set and the historical equipment condition data are inputted into the trained classifier model; the fault diagnosis prediction result outputted by the classifier model and the feature ranking of the feature extraction model are obtained; and the parameters of the feature extraction model are adjusted according to the feature ranking. The invention uses multiple feature extraction models to extract features from data, making fuller and more comprehensive use of data; at the same time, the feedback iterative optimization method is used to adjust the parameters and weight ratios of the combination of multiple models, so that it can be more suitable for fault diagnosis in this scenario. At the same time, the input streaming data is monitored in real time, and "early warning points" are set to improve the efficiency of abnormal data detection.
[0052] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the accompanying drawings:
[0054] Figure 1 A flowchart of a feature extraction optimization method provided in an embodiment of the present application;
[0055] Figure 2 A structural block diagram of a fault prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0056] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0057] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back...), such directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.
[0058] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist, and is not in the flow chart of the feature extraction optimization method required by the present application.
[0059] like Figure 1 As shown, the feature extraction optimization method of the embodiment of the present invention may include the following steps:
[0060] Step S10, obtaining the pre-processed equipment operating condition data;
[0061] The feature extraction optimization method of the embodiment of the present invention is applied to fault diagnosis application scenarios, including fault diagnosis of various types of engineering machinery, vehicles, etc. Before performing the feature extraction optimization method in fault diagnosis of the embodiment of the present invention, it is necessary to upload various types of data of the vehicle or mechanical equipment through sensors on the engineering machinery or vehicle, etc. The various types of data include vibration data, temperature data, sound data, speed data, etc. Later, by extracting features from these data, the extracted features are used to determine whether the equipment has a fault.
[0062] Specifically, when performing data collection applications, data from sensors is collected through the data acquisition module, and its functions mainly include: 1. Data collection capability, where the collected data includes vibration data, temperature data, sound data, and rotation speed data, etc., and the sensor data is collected at a specified frequency; 2. Multi-device collection capability, each data acquisition module can support simultaneous connection with multiple sensors and other devices; 3. Real-time data collection and transmission capability, the data acquisition module can collect data in real time and transmit it in real time.
[0063] When applying data preprocessing, necessary data preprocessing is performed before the data is transmitted to the model collaboration module. Its functions mainly include 1. removing noise caused by environmental interference, etc., and retaining signals related to faults; 2. outlier processing, detecting and processing possible outliers; 3. missing value processing, detecting and processing missing values; 4. performing sample segmentation on the data, etc. Data preprocessing adopts commonly used processing methods in the field, which will not be described here.
[0064] Step 20, matching at least one feature extraction model according to the equipment operating condition data;
[0065] The equipment operating condition data is input into the model collaboration module, which includes at least one feature extraction model. The user can select at least one feature extraction model in the model collaboration module to perform feature extraction processing on the equipment operating condition data as needed. It is also possible to match at least one suitable feature extraction model to the input equipment operating condition data for feature extraction processing.
[0066] The feature extraction models deployed in the model collaboration module include but are not limited to 1. Statistical feature extraction models, statistical feature models based on the time domain, including mean, standard deviation, variance, root amplitude, waveform index, margin index, skewness and kurtosis, etc.; statistical feature models based on the frequency domain, including spectrum energy, spectrum mean, spectrum peak, etc.; 2. Traditional signal processing models, mainly including Fourier transform, wavelet transform, autocorrelation function and spectrum analysis, etc.; 3. Other models, including feature extraction models based on information theory, deep learning models and dimensionality reduction models, etc.
[0067] In this embodiment, one or more combined models can be selected to extract features from data during feature extraction, and a multidimensional feature set can be obtained according to the model weight ratio. ,in Represents the weight ratio of the feature extraction model, middle Represented as the first Features, Indicates that the feature is The model weight ratio is a pre-set parameter or can be customized according to user needs.
[0068] It should be noted that for each feature extraction model, its input and output are clearly defined. Ensure that they can interact effectively and that the output features can provide valuable information for subsequent classifiers. That is, to implement the model integration framework, including inputting raw data into each feature extraction model, training and optimizing each feature extraction model at the same time, ensuring that useful features can be learned, and implementing collaborative work logic. When selecting multiple feature extraction models, it is clear that each of them has unique advantages and characteristics, ensuring that each model can effectively cooperate rather than conflict with each other.
[0069] In addition, in this embodiment, relevant debugging and verification are required when deploying the model collaboration module. This includes verifying the function of the model collaboration module based on offline streaming data. If the function is normal, the offline data is replaced with real-time collected and transmitted streaming data to debug each feature extraction model in the module. If the debugging is correct, the model collaboration module can be deployed for overall verification testing. At the same time, the relevant API is designed in the model collaboration module to import new feature extraction models, and relevant tests are performed after the import. With the help of the scalability of the model collaboration module, it will support the real-time operation of various feature extraction models, and at the same time, it can complete functions such as importing new feature extraction models during real-time operation, allowing the model to continuously learn and adapt to new failure modes during online deployment, showing extremely strong scalability.
[0070] Step 30, obtaining a multidimensional feature set of equipment operating condition data outputted by at least one feature extraction model;
[0071] Obtain real-time equipment operating data of equipment operation, select one or more feature extraction models to extract features from the preprocessed data, and obtain a multidimensional feature set according to the given weight ratio of each model. .
[0072] Specifically, users can configure models according to different needs, select one or more models in the model collaboration module to extract features from the preprocessed data, and obtain a multidimensional feature set according to the given weight ratio of each model, and output it to the clustering module. At the same time, the model collaboration module is updatable, that is, the model parameters can be adjusted in real time.
[0073] Step 40, inputting the multidimensional feature set and historical equipment operating condition data into the trained classifier model;
[0074] In this embodiment, one or more models in the model collaboration module are used to train the previously trained classifier model by outputting a multidimensional feature set based on the feature data extracted from the equipment operating condition data according to the weight of each model. The training method of this classifier model includes:
[0075] (1) Obtaining historical data sets of devices The data set contains labeled data, including equipment operation data, i.e., information on whether the equipment is faulty or not. The mean algorithm calculates the cluster center. Clustering ,in is a fault-free data cluster, is a fault data cluster; a cluster safety area is set, with the cluster center as the dot and a circle with the distance from the data point farthest from the cluster center as the radius as the safety area; (2) historical data sets of the equipment are obtained, and the classifier is trained in a supervised manner to obtain a trained classifier.
[0076] Step 50, obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model;
[0077] Perform clustering processing, assign unlabeled data to the nearest cluster, and obtain pseudo-label output. (1) Obtain the operating data of the equipment in real time. The data is unlabeled data. Calculate the distance from each unlabeled sample to the center of each cluster and assign the data sample to the nearest cluster; (2) Set the data points outside the cluster safety area as "warning points" and use the trained classifier to predict them. Perform operations based on the results of the classifier: If the classification result is normal data, update the cluster safety area. If the classification result is fault data, remove it. (3) Update cluster center: Use all labeled and unlabeled data to recalculate the cluster center. For labeled data, use its true label; for unlabeled data, use the mean of the current cluster to which it belongs to update the value of the cluster center.
[0078] The classifier module receives the multidimensional feature set transmitted by the clustering module and combines it with the historical data set into a mixed data set, performs fault diagnosis on the mixed data set, obtains the importance ranking of each feature of the multidimensional feature set, calculates the contrast loss function, and passes both to the RFE module.
[0079] The classification model in the classifier module is selected by the user according to the actual application scenario. It can be a single classification model or an integrated model composed of multiple classification models. The historical data in the mixed data set uses the clustering results as pseudo labels, and uses the trained classifier to train the multidimensional feature set of the mixed data set to obtain the predicted output. At the same time, the comparison loss function is calculated based on the predicted output and the clustering result. The loss function is as follows: ;in is the temperature parameter, which is used to adjust the uniformity of the probability distribution. is the batch size, represents the cosine similarity, , , Represents the output prediction two-dimensional vector of the model, where vector 1 is the probability of a fault, and vector 2 is the probability of no fault. The output of the clustering model is converted into a vector form using one-hot encoding, that is, when the clustering result is no fault, its vector is [0,1], and when the clustering result is a fault, its vector is [1,0];
[0080] The RFE module receives the importance ranking of the multi-dimensional feature set transmitted by the classifier, the comparison loss function value, and the user input to adjust the parameters of the model collaboration module. At the same time, it ensures that the RFE module can perform instant parameter updates when receiving new samples.
[0081] Before deploying the module, first use the training data set to train it, solve the unknown parameters in the system, and then deploy and test it. After the clustering module adjusts the parameters of the model collaboration module, compare the contrast loss function values before and after the adjustment. If the contrast loss function value after the adjustment meets the predetermined requirements, use the current parameters to continue fault diagnosis; if the contrast loss function value after the adjustment does not meet the predetermined requirements, continue to optimize the model collaboration module until the contrast loss function value meets the predetermined requirements.
[0082] The RFE module is mainly used for recursive feature elimination; the specific steps are as follows:
[0083] Input the existing data set into the model collaboration module to obtain a multi-dimensional feature set;
[0084] Use all the features to train a classifier.
[0085] Evaluate each feature based on the classifier's coefficient, weight, importance and other indicators, and obtain the importance ranking of all features;
[0086] Eliminate the lowest-ranked feature among all feature rankings;
[0087] Pass the feature sorting and other information to the parameter adjustment module for parameter adjustment processing;
[0088] Step 60, adjusting parameters of the feature extraction model according to the feature ranking;
[0089] Evaluate whether the performance of the feature extraction model after adjusting the parameters meets the preset requirements. When the performance of the feature extraction model does not meet the preset requirements, repeat steps 10 to 60.
[0090] The parameter adjustment module uses the fuzzy control algorithm to adjust the parameters, that is, the feature ranking obtained in step five is used to adjust the weight ratio of the feature extraction model in step two.
[0091] The specific steps of fuzzy control are as follows: (1) Define the input as feature ranking and the output as model weight ratio; (2) Fuzzify the ranking and define three fuzzy sets: high ranking: top 30%; medium ranking: between 30% and 70%; low ranking: bottom 30%; (3) Define fuzzy rules: First rule, if the ranking is high, increase the weight of the corresponding feature extraction model; second rule, if the ranking is medium, adjust the weight to remain unchanged; third rule, if the ranking is low, reduce the weight of the corresponding feature extraction model; (4) Fuzzy reasoning, calculate the membership degree corresponding to each feature ranking according to the input feature ranking, and calculate the rule activation degree and synthesize; (5) Defuzzification, define three fuzzy sets with weights of medium, low and high, and their corresponding triangular membership functions are as follows: Low weight: Triangular MF, parameters are ; Medium weight: Triangular MF, parameters are ; High weight: Triangular MF, parameters are ;in ,in is the inverse of the sort, is the hyperparameter of the weight, determined by cross-validation, middle The lowest ranking in the high ranking set is the inverse of the sort. middle The inverse of the ranking of the middle in the high ranking set, middle The highest ranking in the high ranking set is ranked in the reverse order according to the fuzzy output value, and so on. , ; Calculate the weight adjustment rate according to the membership function; (6) Update the weight and continue the operation after step 10. If the number of existing features has reached the preset value or the contrast loss function no longer decreases, stop the iteration; otherwise return to step 10 and continue the iteration.
[0092] The parameter adjustment module can also accept real-time input from users and dynamically adjust the parameters of the model collaboration module based on the real-time input from users. Users can write new parameter adjustment methods into the rule library controlled by the module, or adjust the parameters manually.
[0093] By using the method of this patent invention to extract features and through real-time training of the multi-model collaborative system, adaptive parameters suitable for the working condition can be obtained, and finally a feature extraction method and system for different working conditions can be obtained. At the same time, incremental learning of streaming data is used to optimize the feature extraction model in real time, which greatly improves the universality of the feature extraction system and has extremely strong generalization performance.
[0094] Figure 2This is a structural block diagram of a fault prediction device provided in an embodiment of the present application. Figure 2 As shown, an embodiment of the present application provides a fault prediction device, which may include:
[0095] Memory 210, configured to store instructions; and
[0096] The processor 220 is configured to call instructions from the memory 210 and implement the fault prediction model obtained by the above-mentioned feature extraction optimization method when executing the instructions, and output the device fault type through the fault prediction model.
[0097] An embodiment of the present application also provides a storage medium having instructions stored thereon, the instructions being used to enable a machine to execute the above-mentioned feature extraction optimization method.
[0098] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0099] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0100] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0101] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0102] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0103] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0104] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0105] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0106] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A feature extraction optimization method, characterized in that: The method comprises: Step 10, obtaining the pre-processed equipment operating condition data; Step 20, matching at least one feature extraction model according to the equipment operating condition data; Step 30, obtaining a multidimensional feature set of equipment operating condition data outputted by at least one feature extraction model; Step 40, inputting the multidimensional feature set and historical equipment operating condition data into the trained classifier model; Step 50, obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model; Step 60, adjusting parameters of the feature extraction model according to the feature ranking; Evaluate whether the performance of the feature extraction model after adjusting the parameters meets the preset requirements. When the performance of the feature extraction model does not meet the preset requirements, repeat steps 10 to 60. Wherein, before the step of obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model, it also includes: Performing clustering processing on the multidimensional feature set; Acquire the operation data of the device in real time, where the operation data is unlabeled data, calculate the distance between each unlabeled data and the center of each cluster, and assign the unlabeled data to the nearest cluster; Set data points outside the cluster safety area as warning points; Use the trained classifier to make predictions for data outside the cluster safety zone; If the predicted classification result is normal data, updating the cluster safety area; If the classification result is faulty data, the faulty data is removed; Use the data of all labeled equipment operation data and unlabeled data to recalculate the updated cluster center. The step of evaluating whether the performance of the feature extraction model after adjusting the parameters meets the preset requirements includes: The contrast loss function is calculated based on the fault diagnosis prediction result output by the classifier model and the clustering result, and it is determined whether the contrast loss function continues to meet the preset requirements to evaluate whether the performance of the feature extraction model after adjusting the parameters meets the preset requirements, wherein the contrast loss function is: Where τ is the temperature parameter, which is used to adjust the uniformity of the probability distribution, N is the batch size, Represents cosine similarity, y i ,y j ,y k Represents the output prediction two-dimensional vector of the model. Vector 1 is the probability of predicted failure, and vector 2 is the probability of predicted non-failure. The clustering output is converted into vector form using one-hot encoding. The step of obtaining the fault diagnosis prediction result output by the classifier model and the feature ranking of the feature extraction model includes: Training the multidimensional feature set according to the classifier model, and obtaining an evaluation result of the classifier model evaluating the features of the multidimensional feature set according to the evaluation index; Obtaining importance rankings of all features of the multidimensional feature set; Remove the lowest ranking feature from all feature rankings. Wherein, the step of adjusting parameters of the feature extraction model according to the feature sorting includes: The weight ratio of the feature extraction model is adjusted according to the fuzzy control algorithm, and the weight ratio of the feature extraction model after the parameter adjustment is updated.
2. The feature extraction optimization method according to claim 1, characterized in that: The method further comprises: Get the historical data set D of the device h_data , the historical data set D h_data The data in the table has been marked with equipment operation data; Calculating cluster centers using a k-means algorithm for the historical data set; For the historical data set D h_data Perform cluster division C={C1,C2}, where C1 is a non-fault data cluster and C2 is a fault data cluster; Set the cluster safety area, with the cluster center as the dot and the circle with the distance of the data point farthest from the cluster center as the radius as the safety area.
3. The feature extraction optimization method according to claim 1, characterized in that: The step of adjusting the weight ratio of the feature extraction model according to the fuzzy control algorithm includes: The ranking is fuzzified and three fuzzy sets are defined: high ranking: top 30%; middle ranking: between 30% and 70%; low ranking: bottom 30%; Define fuzzy rules: First rule: if the ranking is high, increase the weight of the corresponding feature extraction model; second rule: if the ranking is medium, adjust the weight to remain unchanged; third rule: if the ranking is low, reduce the weight of the corresponding feature extraction model; Fuzzy reasoning, calculates the membership degree corresponding to each feature ranking according to the input feature ranking, and calculates the rule activation degree and synthesizes it; Defuzzification, define three fuzzy sets with medium weight, low weight, and high weight, and the corresponding triangle membership functions are as follows: Low weight: triangular membership function, parameters are (α1, α2, α3); medium weight: triangular membership function, parameters are (β1, β2, β3); high weight: triangular membership function, parameters are (γ1, γ2, γ3); where where z i is the inverse of the sort, μ p is the hyperparameter of the weight, determined by cross-validation, z in α1 i is the lowest ranking in the high ranking set, z in α2 i is the inverse of the ranking of the middle in the high ranking set, z in α3 i The highest ranking in the high ranking set is ranked inversely according to the fuzzy output value; z in β1 i is the lowest ranking in the ranking set, z in β2 i is the inverse of the ranking of the middle ranking set, z in β3 i The highest ranking in the ranking set is ranked in the reverse order according to the fuzzy output value; z in γ1 i is the lowest ranking in the low ranking set, z in γ2 i is the inverse of the ranking of the middle in the low ranking set, z in γ3 i The highest ranking in the low ranking set is sorted in reverse order according to the fuzzy output value; The weight adjustment rate is calculated according to the triangle membership function and the weight ratio of the feature extraction model is adjusted.
4. A fault prediction method, characterized in that: The method comprises: Obtain equipment operating data; The equipment operating condition data input value is used to obtain a fault prediction model using the feature extraction optimization method according to any one of claims 1 to 3; The equipment fault type is outputted through the fault prediction model.
5. A processor, characterized in that: The method is configured to execute the fault prediction method according to claim 4.
6. A storage medium, characterized in that: The storage medium stores instructions, which are used to enable a machine to execute the fault prediction method according to claim 4.
Citation Information
Patent Citations
Tongue image feature extraction method and system and storage medium
CN115810109A
Fault diagnosis method, system and device for vehicle hub bearing unit and medium
CN116561514A
