CN116089870A - Industrial equipment fault prediction method and device based on meta-learning under small sample condition - Google Patents
Industrial equipment fault prediction method and device based on meta-learning under small sample condition Download PDFInfo
- Publication number
- CN116089870A CN116089870A CN202211622060.8A CN202211622060A CN116089870A CN 116089870 A CN116089870 A CN 116089870A CN 202211622060 A CN202211622060 A CN 202211622060A CN 116089870 A CN116089870 A CN 116089870A
- Authority
- CN
- China
- Prior art keywords
- training
- industrial equipment
- fault prediction
- target domain
- prediction model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The application relates to the technical field of fault prediction, in particular to a method and a device for predicting industrial equipment faults under a small sample condition based on meta-learning, wherein the method comprises the following steps: acquiring time sequence data sets of industrial equipment in a source domain and a target domain; constructing a pseudo task set according to the source domain data set; training a meta-learning model in a source domain by utilizing a pseudo task set, and adjusting in a target domain to obtain a fault prediction model suitable for the target domain, wherein the learning rate, training rounds and training sample number of the target domain are all small Yu Yuanyu; and predicting a failure prediction value of the industrial equipment by using a failure prediction model of the target domain, judging that the industrial equipment fails if the failure prediction value is larger than a preset threshold value, otherwise judging that the industrial equipment is normal. Therefore, the problems that in the related art, the accuracy of the depth model obtained by training is low and the failure prediction requirement cannot be met due to the fact that failure samples which can be extracted from industrial equipment are fewer are solved.
Description
Technical Field
The application relates to the technical field of data-driven industrial equipment fault prediction, in particular to an industrial equipment fault prediction method and device based on meta-learning under a small sample condition.
Background
The current fault prediction methods for industrial equipment are mainly divided into three types, namely a fault prediction method based on a physical model, a fault prediction method based on traditional machine learning and a fault prediction method based on deep learning, wherein the fault prediction based on the physical model is performed by modeling and analyzing the physical state of the equipment, and the fault prediction is performed by visual change of the physical state; the traditional machine learning-based method is to fit a functional relation between time sequence data and fault occurrence time through some simple mathematical modeling or neural network structure, and comprises a support vector machine, a Monte Carlo method and the like; due to the improvement of computer power and the rapid development of artificial intelligence technology, a fault prediction method based on deep learning is rapidly developed and has remarkable achievement.
In the related art, a large number of fault samples are required to be trained to obtain an effective depth model in a fault prediction method based on deep learning, but in actual industry, the fault samples which can be acquired are fewer because the equipment is not always maintained until the equipment fails and is maintained after the equipment runs for a period of time, and in addition, the life cycle of part of industrial equipment is always long, so that enough fault samples are not available.
While the present approach can cope with the situation of few failed samples, such as accelerating the degradation process in a laboratory by extreme loads or using components made of fragile materials and simulating the conventional degradation process by applying exponential failure, both approaches alleviate the situation of insufficient real samples to some extent, the realistic working conditions and failure modes cannot be perfectly simulated, resulting in certain errors, making the results inaccurate.
Disclosure of Invention
The application provides an industrial equipment fault prediction method, an industrial equipment fault prediction device, electronic equipment and a storage medium, and aims to solve the problems that in the related art, due to the fact that a fault sample which can be extracted from industrial equipment is less, the accuracy of a depth model obtained through training is low, and the requirements of fault prediction cannot be met.
An embodiment of a first aspect of the present application provides an industrial equipment failure prediction method, including the following steps: acquiring time sequence data sets of industrial equipment in a source domain and a target domain; constructing a pseudo task data set according to the time sequence data set of the source domain, and performing meta-learning training on the source domain by utilizing the pseudo task data set to obtain a fault prediction model applicable to the source domain; the fault prediction model is adjusted by utilizing the time sequence data set of the target domain to obtain a fault prediction model suitable for the target domain, wherein the learning rate of the target domain during training is smaller than that of the source domain during training, the training round of the target domain during training is smaller than that of the source domain during training, and the number of training samples of the target domain is smaller than that of the source domain; and predicting a fault prediction value of the industrial equipment by using the fault prediction model of the target domain, judging that the industrial equipment is faulty if the fault prediction value is larger than a preset threshold value, and otherwise judging that the industrial equipment is normal.
Optionally, the performing meta-learning training on the source domain by using the pseudo task data set to obtain a fault prediction model applicable to the source domain includes: randomly selecting one or more training tasks from the pseudo-task data set; performing meta-learning internal updating on the constructed fault prediction model according to the one or more training tasks, and calculating a training loss value of each training task; and carrying out external updating on the parameters of the fault prediction model according to the comprehensive loss value of the training loss value of each training task, taking the parameters after external updating as final model parameters of one-time training, and stopping iterative training until the training completion condition of the fault prediction model is met, so as to obtain the fault prediction model applicable to the source domain.
Optionally, the adjusting the fault prediction model by using the time series data set of the target domain to obtain a fault prediction model applicable to the target domain includes: and adjusting parameters of a fault prediction model of the source domain according to samples of the time sequence data set so as to migrate the fault prediction model from the source domain to the target domain, and obtaining the fault prediction model applicable to the target domain after parameter adjustment is completed.
Optionally, the constructing a pseudo task data set according to the time series data set of the source domain includes: randomly selecting fault data from the time sequence data set as a test set of training tasks; and calculating similarity measurement of the test sample in the test set and each time sequence in the time sequence data set, selecting a time sequence with a preset similarity measurement larger than the preset measurement as a training set of the training task, and constructing the pseudo task data set based on the test set and the training set.
Optionally, before constructing the pseudo task data set according to the time series data set of the source domain, the method further comprises: and carrying out normalization processing and data enhancement processing on the time sequence data of the target domain, and adding Gaussian noise or uniform noise in different sensor data dimensions to obtain the processed time sequence data.
An embodiment of a second aspect of the present application provides an industrial equipment failure prediction method, including the steps of: acquiring operation data of industrial equipment; inputting the operation data into a trained fault prediction model, and outputting a fault prediction value of the industrial equipment, wherein the fault prediction model is obtained by training the industrial equipment on the basis of a time sequence data set of a source domain and a target domain; and if the fault prediction value is larger than a preset threshold value, judging that the industrial equipment is faulty, otherwise, judging that the industrial equipment is normal.
An embodiment of a third aspect of the present application provides an industrial equipment failure prediction apparatus, including: the first acquisition module is used for acquiring a time sequence data set of the industrial equipment in a source domain and a target domain; the construction module is used for constructing a pseudo task data set according to the time sequence data set of the source domain, and performing meta learning training on the source domain by utilizing the pseudo task data set to obtain a fault prediction model applicable to the source domain; the adjusting module is used for adjusting the fault prediction model by utilizing the time sequence data set of the target domain to obtain a fault prediction model suitable for the target domain, wherein the learning rate of the target domain during training is smaller than that of the source domain during training, the training round of the target domain during training is smaller than that of the source domain during training, and the number of training samples of the target domain is smaller than that of the source domain; and the prediction module is used for predicting the failure prediction value of the industrial equipment by utilizing the failure prediction model of the target domain, judging that the industrial equipment fails if the failure prediction value is larger than a preset threshold value, and judging that the industrial equipment is normal if the failure prediction value is not larger than the preset threshold value.
An embodiment of a fourth aspect of the present application provides an industrial equipment failure prediction apparatus, including: the second acquisition module is used for acquiring the operation data of the industrial equipment; the input/output module is used for inputting the operation data into a trained fault prediction model and outputting a fault prediction value of the industrial equipment, wherein the fault prediction model is obtained by training the industrial equipment on the basis of a time sequence data set of a source domain and a target domain; and the judging module is used for judging that the industrial equipment is faulty if the fault prediction value is larger than a preset threshold value, and judging that the industrial equipment is normal if the fault prediction value is not smaller than the preset threshold value.
An embodiment of a fifth aspect of the present application provides an electronic device, including: the system comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the industrial equipment fault prediction method according to the embodiment.
An embodiment of a sixth aspect of the present application provides a computer-readable storage medium having stored thereon a computer program that is executed by a processor for implementing the industrial equipment failure prediction method as described in the above embodiment.
Therefore, the application has at least the following beneficial effects:
(1) The model trained by the meta-learning training method is not subjected to parameter optimization aiming at a certain task, but is applicable to various tasks; therefore, the model parameters trained by the method can be quickly adapted to new tasks under the fine adjustment of a small number of samples, and industrial equipment fault prediction under the condition of the small samples is completed.
(2) The embodiment of the application uses an advanced meta-learning training method and a pseudo task set construction method, so that the model can be quickly adapted to the target domain working condition, and the fault prediction accuracy of the depth model under the condition of a small sample is greatly improved.
(3) The embodiment of the application does not modify the deep neural network structure, is theoretically suitable for most gradient descent algorithms, can be seamlessly combined with the traditional fault prediction model, and improves the application range of the embodiment of the application.
Additional aspects and advantages of the application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application.
Drawings
The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow chart of a method of industrial equipment fault prediction provided in accordance with one embodiment of the present application;
FIG. 2 is a flowchart of training and use of an industrial equipment failure prediction method based on meta-learning under small sample conditions provided in accordance with an embodiment of the present application;
FIG. 3 is a flow chart of a method of industrial equipment fault prediction provided in accordance with another embodiment of the present application;
FIG. 4 is a block schematic diagram of an industrial equipment failure prediction apparatus provided according to one embodiment of the present application;
FIG. 5 is a block schematic diagram of an industrial equipment failure prediction apparatus provided in accordance with another embodiment of the present application;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and intended for the purpose of explaining the present application and are not to be construed as limiting the present application.
Industrial equipment fault prediction methods, devices, electronic equipment and storage media according to embodiments of the present application are described below with reference to the accompanying drawings. Aiming at the problems that in the related art mentioned in the background art, the accuracy of a depth model obtained by training is low and the failure prediction requirement cannot be met due to fewer failure samples which can be extracted from industrial equipment, and the like, the application provides an industrial equipment failure prediction method, wherein a pseudo task set is constructed according to a source domain data set by acquiring a time sequence data set of industrial equipment in a source domain and a target domain; the training of the meta-learning model is carried out in the source domain by utilizing the pseudo task set, the adjustment is carried out in the target domain, the fault prediction model suitable for the target domain is obtained, the fault prediction value of the industrial equipment is predicted by utilizing the fault test model of the target domain, if the fault prediction value is larger than a certain threshold value, the industrial equipment is judged to be faulty, otherwise, the equipment is judged to be normal, the training of the equipment fault prediction model under the condition of fewer fault samples can be met, meanwhile, the method is not limited by a neural network structure, the current vast majority of gradient descent algorithms can be met, the fault prediction of the target equipment under the condition of small samples is realized, the fault prediction accuracy of the depth model under the condition of small samples is greatly improved, and the application range is improved. Therefore, the problems that in the related art, the accuracy of the depth model obtained by training is low and the failure prediction requirement cannot be met due to the fact that failure samples which can be extracted from industrial equipment are fewer are solved.
Specifically, fig. 1 is a schematic flow chart of an industrial equipment fault prediction method provided in an embodiment of the present application.
As shown in fig. 1, the industrial equipment failure prediction method includes the steps of:
in step S101, a time series dataset of the industrial device in a source domain and a target domain is acquired.
Wherein, the source domain represents the labeled domain capable of training the model; the target field means a field having no label or only a small number of labels, and is not particularly limited herein.
The time series data set may be all data from the life cycle of the industrial equipment to the moment of the sensor recording, such as data after related preprocessing, which can reflect the state of the industrial equipment, and time series data after preprocessing, such as equipment working condition recording data, operation state data, and the like, recorded by the sensor, such as flow sensor data, pressure sensor data, temperature sensor data, and the like, which are used for describing the situation that the phenomenon changes with time, and are not particularly limited herein.
It can be appreciated that in the embodiment of the application, the time series data sets of the industrial equipment in the source domain and the target domain are acquired through different sensors, so that the data sets can be trained to obtain the fault prediction data model later.
In step S102, a pseudo task data set is constructed according to the time-series data set of the source domain, and meta learning training is performed in the source domain by using the pseudo task data set, so as to obtain a fault prediction model suitable for the source domain.
The pseudo task data set may match the most similar time series data for each tag sample, so that the tag approximates to a real note, which is not specifically limited herein.
The meta learning training can enable the model to acquire the learning parameter adjustment capability, so that the model can quickly learn a new task on the basis of acquiring the existing knowledge, and the meta learning training is not particularly limited.
It can be understood that the time sequence data set of the source domain in the embodiment of the application constructs a pseudo task data set, and performs meta-learning training on the source domain by using the pseudo task data set to obtain a fault prediction model suitable for the source domain, so that the meta-learning model trained on the source domain can be conveniently transferred to the target domain subsequently.
In an embodiment of the present application, before constructing the pseudo task data set according to the time series data set of the source domain, the method further includes: and carrying out normalization processing and data enhancement processing on the time series data of the target domain, and adding Gaussian noise or uniform noise in different sensor data dimensions to obtain the processed time series data.
The normalization processing may normalize the numerical range of the physical quantity of different scales to between 0 and 1, and the calculation formula of the normalization processing is as follows:
wherein the maximum value in the characteristic latitude is x max The minimum value is x min The normalized data is x.
The data enhancement process may be to randomly select a proportion between-10% and 10% based on the current sensor value, multiply the proportion by the current sensor value, and apply the multiplied proportion to the current sensor value, which is not specifically limited herein.
It can be understood that, in the embodiment of the application, the time series data of the target domain is subjected to normalization processing and data enhancement processing, so that the convergence speed of the neural network can be increased, the training speed of the network is increased, and the fault prediction accuracy of the deep neural network is improved; and Gaussian noise or uniform noise is added in different sensor data dimensions to obtain processed time series data, so that the generalization capability of the fault prediction model is improved.
In an embodiment of the present application, constructing a pseudo task data set from a time-series data set of a source domain includes: randomly selecting fault data from the time sequence data set as a test set of training tasks; and calculating the similarity measurement of each time sequence in the test sample and the time sequence data set in the test set, selecting a time sequence with a preset similarity measurement larger than the preset measurement as a training set of the training task, and constructing a pseudo task data set based on the test set and the training set.
The similarity measure may be a euclidean distance similarity measure or a DTW similarity measure, which may be selected according to the actual requirement of the user, and is not specifically limited herein.
The Euclidean distance similarity measurement and calculation mode is as follows:
wherein the method comprises the steps of
The first moment of the ith time sequence, likewise->
The first time of the j-th time series is indicated.
The calculation mode of the DTW similarity measurement is as follows:
wherein the method comprises the steps of
The p-th moment of the ith time sequence, similarly->
The (q) th moment of the j-th time sequence, is shown>
Is a transformation function for time series of different lengths, < >>
Is a normalization factor and can be expressed as the sum of the lengths of the two time sequences.
The preset similarity measure may be set with a plurality of similarity measures according to the user requirement, which is not limited herein.
The time sequence of the preset measurement may be a time sequence of a measurement preset by a user, which is not specifically limited herein.
It can be understood that, in the embodiment of the present application, certain fault data is randomly selected from the time series data set as a test set of the training task, similarity measurement between a test sample in the test set and each time series in the time series data set is calculated, a time series with a relative degree greater than a certain value is selected as a training set of the training task, and a pseudo task data set is constructed based on the test set and the training set, so that a meta-learning model is trained in a source domain and is migrated to a target domain in the following process.
In step S103, the fault prediction model is adjusted by using the time sequence data set of the target domain, so as to obtain a fault prediction model suitable for the target domain, wherein the learning rate of the target domain during training is smaller than that of the source domain during training, the training round of the target domain during training is smaller than that of the source domain during training, and the number of training samples of the target domain is smaller than that of the source domain.
The fault prediction model may be any deep neural network model capable of completing fault prediction, and is not specifically limited herein.
The learning rate may be an important super parameter in supervised learning and deep learning, which determines whether the objective function can converge to a local minimum and when it converges to a minimum, and the appropriate learning rate enables the objective function to converge to a local minimum in an appropriate time, which is not particularly limited herein.
It can be understood that, in the embodiment of the application, the time sequence data set of the target domain is utilized to adjust the fault prediction model, so that the fault prediction model applicable to the target domain is obtained, the fault prediction of the target device under the condition of a small sample can be realized, and the fault prediction accuracy of the depth model under the condition of the small sample is improved.
It should be noted that, the adjustment of the fault prediction model by using the time sequence data set of the target domain means that in deep learning, the model is firstly trained in the source domain by using a large amount of data, and when the model is migrated to the target domain, because the model parameters are not expected to change too much, a small amount of samples of the target domain are used, the model parameters are simply updated, and the fault prediction accuracy of the fault prediction model under the condition of small samples is improved.
In the embodiment of the application, meta-learning training is performed in a source domain by using a pseudo task data set to obtain a fault prediction model applicable to the source domain, including: randomly selecting one or more training tasks from the pseudo-task data set; performing meta-learning internal updating on the constructed fault prediction model according to one or more training tasks, and calculating a training loss value of each training task; and carrying out external updating on the parameters of the fault prediction model according to the comprehensive loss value of the training loss value of each training task, taking the parameters after external updating as final model parameters of one-time training, and stopping iterative training until the training completion condition of the fault prediction model is met, so as to obtain the fault prediction model applicable to the source domain.
It can be understood that in the embodiment of the present application, training tasks are randomly selected from the pseudo task data set, meta-learning internal update is performed on the constructed fault prediction model, a training loss value of each training task is calculated, a comprehensive loss value is calculated according to the training loss value of each training task, and external update is performed on parameters of the fault prediction model, and the calculated comprehensive loss value is used as final model parameters of one-time training until training completion conditions of the fault prediction model are met, and a traditional deep learning training mode is modified into two training stages of internal update and external update, so that a current vast majority of gradient descent models can be met, and fault prediction of target equipment under a small sample condition is realized.
In the embodiment of the present application, the adjusting the fault prediction model by using the time sequence data set of the target domain to obtain the fault prediction model applicable to the target domain includes: and adjusting parameters of the fault prediction model of the source domain according to samples of the time sequence data set so as to migrate the fault prediction model from the source domain to the target domain, and obtaining the fault prediction model applicable to the target domain after the parameter adjustment is completed.
It can be understood that in the embodiment of the present application, parameters of a fault prediction model of a source domain are adjusted according to samples of a time sequence data set, the fault prediction model is migrated from the source domain to a target domain, and after parameter adjustment is completed, a fault prediction model applicable to the target domain is obtained, so that the model can be quickly adapted to working conditions of the target domain, and the fault prediction accuracy of a depth model under the condition of a small sample is greatly improved.
In step S104, the failure prediction value of the industrial equipment is predicted by using the failure prediction model of the target domain, if the failure prediction value is greater than the preset threshold, the industrial equipment is determined to be failed, otherwise, the industrial equipment is determined to be normal.
The preset threshold may be a threshold set by a user, and may be set according to actual situations, which is not specifically limited herein.
It can be understood that, in the embodiment of the application, the fault prediction value of the industrial equipment is predicted by using the fault test model of the target domain, if the fault prediction value is greater than a certain value, the industrial equipment is judged to be faulty, otherwise, the equipment is judged to be normal, and the accuracy of predicting the industrial equipment fault is improved according to the fault test model.
According to the industrial equipment fault prediction method provided by the embodiment of the application, a pseudo task set is constructed according to a source domain data set by acquiring the time sequence data sets of industrial equipment in a source domain and a target domain; the training of the meta-learning model is carried out in the source domain by utilizing the pseudo task set, the adjustment is carried out in the target domain, the fault prediction model suitable for the target domain is obtained, the fault prediction value of the industrial equipment is predicted by utilizing the fault test model of the target domain, if the fault prediction value is larger than a certain threshold value, the industrial equipment is judged to be faulty, otherwise, the equipment is judged to be normal, the training of the equipment fault prediction model under the condition of fewer fault samples can be met, meanwhile, the method is not limited by a neural network structure, the current vast majority of gradient descent algorithms can be met, the fault prediction of the target equipment under the condition of small samples is realized, the fault prediction accuracy of the depth model under the condition of small samples is greatly improved, and the application range is improved. Therefore, the problems that in the related art, the accuracy of the depth model obtained by training is low and the failure prediction requirement cannot be met due to the fact that failure samples which can be extracted from industrial equipment are fewer are solved.
The industrial equipment fault prediction method is described in detail as follows:
step S1: typical industrial equipment sensor data is acquired and a time series dataset is constructed in a source domain and a target domain under the health state of the target equipment.
Wherein the time series data set comprises all data of the target device from the beginning of the life cycle to the moment of sensor recording.
Step S2: and screening the sensor dimension capable of reflecting the health state degradation of the target equipment from the time series data set by using the characteristic judgment standard.
And screening out the preprocessed data which can reflect the state of the target equipment, the equipment working condition record data, the operation state data and other preprocessed time sequence data from the data such as the flow sensor data, the pressure sensor data, the temperature sensor data and the like.
Step S3: performing data preprocessing on the screened sensor feature dimensions to obtain feature vectors, wherein the data preprocessing can comprise: data normalization, data enhancement, etc.
The preprocessing aims to normalize the numerical ranges of physical quantities with different scales to be between 0 and 1, so that the convergence speed of the neural network can be increased, the training speed of the network can be increased, and the fault prediction accuracy of the deep neural network can be improved.
Specifically, the embodiment of the application uses a Max-Min normalization mode, and the calculation formula of the normalization process is as follows:
wherein the maximum value in the characteristic latitude is x max The minimum value is x min The normalized data is x.
In addition, for the collected time series data, gaussian noise or uniform noise with proper amplitude can be added to different sensor data dimensions, so that the generalization capability of the fault prediction model is improved. The method is characterized in that a certain amplitude Gaussian noise is generated before time series data normalization, the Gaussian noise is added to the data of the sensor, and the method for adding uniform noise is to randomly select a proportion between-10% and 10% on the basis of the value of the current sensor before data normalization, multiply the proportion by the current sensor value and then add the proportion to the current sensor value.
Step S4: and carrying out fault prediction on the target equipment according to the feature vector and the depth fault prediction model.
The premise of carrying out fault prediction on the target equipment is to obtain a fault prediction model suitable for a target domain, wherein a pseudo task set can be constructed by a time sequence similarity matching method, and then a corresponding prediction fault model is obtained by carrying out learning and test training on a pseudo task data set, so that the task quantity requirement of a meta-learning training model is met.
In the method, a meta-learning model needs to be trained in a source domain and migrated to a target domain, so that a pseudo task set needs to be built in the source domain and the target domain.
The concrete construction mode is as follows: randomly selecting one piece of fault data from source domains with rich fault samples as a test set test of task n n Then selecting TopK most similar time sequences in a source domain by using Euclidean distance similarity measure or DTW similarity measure as a training set train of the task n n . Task n may thus be denoted as test n And train n Is expressed as a task n 。
The Euclidean distance similarity measurement and calculation mode is as follows:
wherein the method comprises the steps of
The first moment of the ith time sequence, likewise->
The first time of the j-th time series is indicated. />
The calculation mode of the DTW similarity measurement is as follows:
wherein the method comprises the steps of
The p-th moment of the ith time sequence, similarly->
The (q) th moment of the j-th time sequence, is shown>
Is a transformation function for time series of different lengths, < >>
Is a normalization factor and can be expressed as the sum of the lengths of the two time sequences.
In the target domain, there are a small number of failure samples and by having several time sequences that need to be failure predicted, these time sequences are untagged. At this time, in a similar manner to the construction of the pseudo task set in the source domain, K most similar time sequences are matched in the target domain for each unlabeled sample, and after the model is trimmed by the matched time sequences, fault prediction is performed on the unlabeled sample.
The fault prediction model may be any deep neural network model capable of completing fault prediction, and different data processing modes may be required for different neural network models, and a transform network will be specifically described below as an example:
the data are sorted into b multiplied by q multiplied by dim dimensions, wherein dim is a sensor characteristic dimension, seq is expressed as the data length of a sensor record, and b is expressed as batch (batch processing), so that the requirement of parallelization calculation of a neural network is met; the seq length in one batch should be kept consistent, so that 0 can be complemented for a shorter time sequence, thereby ensuring the consistency of the sequence length; the transducer Model is represented as a Model, which can be virtually any depth Model that can accomplish deep neural network training.
While the training of the fault prediction model is divided into two stepsSegments, i.e., inner-Update and Outer-Update. The gradient Update mode in the Inner-Update stage is consistent with the traditional deep learning training mode, for example, a task is selected randomly n And training a model by using the task, wherein the model parameter updating mode is as follows:
only one gradient descent is used here for the sake of presentation simplicity, in practice multiple gradient descent may be used in the Inner-update, then the updated model is used in the task n Test set test of (2) n Testing to obtain the loss of the task
The above expressions are all the cases on one task, but in practice, the Batch tasks are randomly selected to perform the Inner-update, and after the loss of the tasks is obtained, the Outer-update is performed, and the calculation formula is as follows:
specifically, in one embodiment of the present application, pseudo codes trained based on the device failure prediction model under meta-learning small sample conditions are shown in table 1:
TABLE 1
The embodiment of the application uses the Root Mean Square Error (RMSE) which is a commonly used evaluation function in the field of health evaluation and fault prediction to evaluate the prediction accuracy of the model, and the calculation mode is as follows:
where m is the number of samples, y t Is a true label, y * To predict tags.
Specifically, as shown in fig. 2, the fault prediction method in the embodiment of the present application specifically includes the following two stages:
(1) Meta learning stage: for each time sequence in the source domain, it is taken as a test set test of task n n Matching K most similar time sequences for each time sequence in a source domain according to a time similarity matching method, and taking the K most similar time sequences as a training set train of a task n n The method comprises the steps of carrying out a first treatment on the surface of the And then randomly selecting a training set of the batch pseudo tasks to perform internal update of meta-learning, wherein the stage does not directly change the parameters of the model, and performs external update after calculating the loss of each pseudo task under the model parameters at the moment, and the parameters after external update are used as final model parameters of one-time training, so that the meta-learning stage is completed after multiple times of training.
(2) Meta-test stage: k similar samples are matched from a small number of samples of a target source in a time similarity matching mode aiming at each time sequence sample needing fault prediction, a training set of the sample is formed, parameters of the model are finely adjusted by using the training set, and then fault prediction is carried out on the sample.
The method and the device are applicable to target industrial equipment with few fault samples, accurate fault prediction is completed, meanwhile, the deep neural network structure is not modified, the method and the device are theoretically applicable to most gradient descent algorithms, and prediction of the equipment health state under the condition of small samples can be achieved.
It should be noted that, in the above embodiment, the fault prediction is performed on the industrial equipment by constructing the fault prediction model in an online state, so as to implement the fault prediction of the target equipment under the condition of a small sample; the following embodiments are specific to performing fault prediction on industrial equipment by using a fault prediction model in an offline state:
fig. 3 is a flow chart of an industrial equipment failure prediction method of an embodiment of the present application.
As shown in fig. 3, the industrial equipment failure prediction method includes the steps of:
in step S201, operation data of an industrial device is acquired.
It can be appreciated that, in the embodiment of the present application, the operation data of the industrial device is acquired through the acquisition device, so as to facilitate the subsequent judgment of whether the device is normal.
In step S202, the operation data is input into a trained failure prediction model, and a failure prediction value of the industrial equipment is output, wherein the failure prediction model is obtained by training based on a time series data set of the industrial equipment in a source domain and a target domain.
It can be appreciated that the embodiment of the application inputs the operation data into the trained fault prediction model, and outputs the fault prediction value of the industrial equipment, so as to judge whether the industrial equipment is faulty or not according to the fault prediction value.
In step S203, if the failure prediction value is greater than the preset threshold, the industrial equipment is determined to be failed, otherwise, the industrial equipment is determined to be normal.
It can be understood that in the embodiment of the present application, if the fault prediction value is greater than the set threshold, the fault device is determined to be faulty, otherwise, it is determined that the industrial device can normally operate, so that the prediction of the device health status under the condition of a small sample can be implemented.
According to the industrial equipment fault prediction method provided by the embodiment of the application, the operation data of the industrial equipment is acquired, the operation data is input into the trained fault prediction model, the fault prediction value of the industrial equipment is output, if the fault prediction value is larger than the set threshold value, the fault of the industrial equipment is judged, otherwise, the industrial equipment can be normally operated, and therefore the prediction of the equipment health state under the condition of a small sample can be achieved.
Next, an industrial equipment failure prediction apparatus according to an embodiment of the present application will be described with reference to the accompanying drawings.
Fig. 4 is a block schematic diagram of an industrial equipment failure prediction apparatus according to an embodiment of the present application.
As shown in fig. 4, the industrial equipment failure prediction apparatus 20 includes: a first acquisition module 210, a construction module 220, a training module 230, and a prediction module 240.
Wherein, the first obtaining module 210 is configured to obtain a time series data set of the industrial device in a source domain and a target domain; the construction module 220 is configured to construct a pseudo task data set according to the time sequence data set of the source domain, and perform meta-learning training on the source domain by using the pseudo task data set to obtain a fault prediction model applicable to the source domain; the adjustment module 230 is configured to adjust the fault prediction model by using the time sequence data set of the target domain to obtain a fault prediction model applicable to the target domain, where the learning rate of the target domain during training is smaller than the learning rate of the source domain during training, the training round of the target domain during training is smaller than the training round of the source domain during training, and the number of training samples of the target domain is smaller than the number of training samples of the source domain; the prediction module 240 is configured to predict a failure prediction value of the industrial equipment by using a failure prediction model of the target domain, and if the failure prediction value is greater than a preset threshold, determine that the industrial equipment is failed, otherwise determine that the industrial equipment is normal.
It should be noted that the foregoing explanation of the embodiment of the industrial equipment fault prediction method is also applicable to the industrial equipment fault prediction device of this embodiment, and will not be repeated herein.
According to the industrial equipment fault prediction device provided by the embodiment of the application, a pseudo task set is constructed according to a source domain data set by acquiring the time sequence data sets of industrial equipment in a source domain and a target domain; the training of the meta-learning model is carried out in the source domain by utilizing the pseudo task set, the adjustment is carried out in the target domain, the fault prediction model suitable for the target domain is obtained, the fault prediction value of the industrial equipment is predicted by utilizing the fault test model of the target domain, if the fault prediction value is larger than a certain threshold value, the industrial equipment is judged to be faulty, otherwise, the equipment is judged to be normal, the training of the equipment fault prediction model under the condition of fewer fault samples can be met, meanwhile, the method is not limited by a neural network structure, the current vast majority of gradient descent algorithms can be met, the fault prediction of the target equipment under the condition of small samples is realized, the fault prediction accuracy of the depth model under the condition of small samples is greatly improved, and the application range is improved. Therefore, the problems that in the related art, the accuracy of the depth model obtained by training is low and the failure prediction requirement cannot be met due to the fact that failure samples which can be extracted from industrial equipment are fewer are solved.
Next, an industrial equipment failure prediction apparatus according to an embodiment of the present application will be described with reference to the accompanying drawings.
Fig. 5 is a block schematic diagram of an industrial equipment failure prediction apparatus according to an embodiment of the present application.
As shown in fig. 5, the industrial equipment failure prediction apparatus 30 includes: a second acquisition module 310, an input output module 320, and a determination module 330.
Wherein, the second obtaining module 310 is configured to obtain operation data of the industrial device; the input/output module 320 is configured to input the operation data into a trained failure prediction model, and output a failure prediction value of the industrial device, where the failure prediction model is obtained by training the industrial device based on a time series data set of a source domain and a target domain; the determining module 330 is configured to determine that the industrial equipment is faulty if the fault prediction value is greater than a preset threshold, and determine that the industrial equipment is normal otherwise.
It should be noted that the foregoing explanation of the embodiment of the industrial equipment fault prediction method is also applicable to the industrial equipment fault prediction device of this embodiment, and will not be repeated herein.
According to the industrial equipment fault prediction method provided by the embodiment of the application, the operation data of the industrial equipment is acquired, the operation data is input into the trained fault prediction model, the fault prediction value of the industrial equipment is output, if the fault prediction value is larger than the set threshold value, the fault of the industrial equipment is judged, otherwise, the industrial equipment can be normally operated, and therefore the prediction of the equipment health state under the condition of a small sample can be achieved.
Fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device may include:
a memory 601, a processor 602, and a computer program stored on the memory 601 and executable on the processor 602.
The processor 602 implements the industrial equipment failure prediction method provided in the above embodiments when executing a program.
Further, the electronic device further includes:
a communication interface 603 for communication between the memory 601 and the processor 602.
A memory 601 for storing a computer program executable on the processor 602.
The memory 601 may include a high-speed RAM (Random Access Memory ) memory, and may also include a nonvolatile memory, such as at least one disk memory.
If the memory 601, the processor 602, and the communication interface 603 are implemented independently, the communication interface 603, the memory 601, and the processor 602 may be connected to each other through a bus and perform communication with each other. The bus may be an ISA (Industry Standard Architecture ) bus, a PCI (Peripheral Component, external device interconnect) bus, or EISA (Extended Industry Standard Architecture ) bus, among others. The buses may be divided into address buses, data buses, control buses, etc. For ease of illustration, only one thick line is shown in fig. 6, but not only one bus or one type of bus.
Alternatively, in a specific implementation, if the memory 601, the processor 602, and the communication interface 603 are integrated on a chip, the memory 601, the processor 602, and the communication interface 603 may perform communication with each other through internal interfaces.
The processor 602 may be a CPU (Central Processing Unit ) or ASIC (Application Specific Integrated Circuit, application specific integrated circuit) or one or more integrated circuits configured to implement embodiments of the present application.
The embodiments of the present application also provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the industrial equipment failure prediction method as above.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. 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 N embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "N" is at least two, such as two, three, etc., unless explicitly defined otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more N executable instructions for implementing specific logical functions or steps of the process, and further implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present application.
It is to be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the N steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As with the other embodiments, if implemented in hardware, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable gate arrays, field programmable gate arrays, and the like.
Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.
Claims (10)
1. A method for predicting failure of an industrial device, comprising the steps of:
acquiring time sequence data sets of industrial equipment in a source domain and a target domain;
constructing a pseudo task data set according to the time sequence data set of the source domain, and performing meta-learning training on the source domain by utilizing the pseudo task data set to obtain a fault prediction model applicable to the source domain;
the fault prediction model is adjusted by utilizing the time sequence data set of the target domain to obtain a fault prediction model suitable for the target domain, wherein the learning rate of the target domain during training is smaller than that of the source domain during training, the training round of the target domain during training is smaller than that of the source domain during training, and the number of training samples of the target domain is smaller than that of the source domain;
And predicting a fault prediction value of the industrial equipment by using the fault prediction model of the target domain, judging that the industrial equipment is faulty if the fault prediction value is larger than a preset threshold value, and otherwise judging that the industrial equipment is normal.
2. The method according to claim 1, wherein the performing meta-learning training in the source domain using the pseudo task data set to obtain a fault prediction model applicable to the source domain includes:
randomly selecting one or more training tasks from the pseudo-task data set;
performing meta-learning internal updating on the constructed fault prediction model according to the one or more training tasks, and calculating a training loss value of each training task;
and carrying out external updating on the parameters of the fault prediction model according to the comprehensive loss value of the training loss value of each training task, taking the parameters after external updating as final model parameters of one-time training, and stopping iterative training until the training completion condition of the fault prediction model is met, so as to obtain the fault prediction model applicable to the source domain.
3. The method of claim 1, wherein said adjusting the fault prediction model using the time series dataset of the target domain to obtain a fault prediction model applicable to the target domain comprises:
And adjusting parameters of a fault prediction model of the source domain according to samples of the time sequence data set so as to migrate the fault prediction model from the source domain to the target domain, and obtaining the fault prediction model applicable to the target domain after parameter adjustment is completed.
4. The method of claim 1, wherein said constructing a pseudo-task data set from a time-series data set of said source domain comprises:
randomly selecting fault data from the time sequence data set as a test set of training tasks;
and calculating similarity measurement of the test sample in the test set and each time sequence in the time sequence data set, selecting a time sequence with a preset similarity measurement larger than the preset measurement as a training set of the training task, and constructing the pseudo task data set based on the test set and the training set.
5. The method of claim 1, further comprising, prior to constructing a pseudo-task dataset from the time-series dataset of source domains:
and carrying out normalization processing and data enhancement processing on the time sequence data of the target domain, and adding Gaussian noise or uniform noise in different sensor data dimensions to obtain the processed time sequence data.
6. A method for predicting failure of an industrial device, comprising the steps of:
acquiring operation data of industrial equipment;
inputting the operation data into a trained fault prediction model, and outputting a fault prediction value of the industrial equipment, wherein the fault prediction model is obtained by training the industrial equipment on the basis of a time sequence data set of a source domain and a target domain;
and if the fault prediction value is larger than a preset threshold value, judging that the industrial equipment is faulty, otherwise, judging that the industrial equipment is normal.
7. An industrial equipment failure prediction apparatus, comprising:
the first acquisition module is used for acquiring a time sequence data set of the industrial equipment in a source domain and a target domain;
the construction module is used for constructing a pseudo task data set according to the time sequence data set of the source domain, and performing meta learning training on the source domain by utilizing the pseudo task data set to obtain a fault prediction model applicable to the source domain;
the adjusting module is used for adjusting the fault prediction model by utilizing the time sequence data set of the target domain to obtain a fault prediction model suitable for the target domain, wherein the learning rate of the target domain during training is smaller than that of the source domain during training, the training round of the target domain during training is smaller than that of the source domain during training, and the number of training samples of the target domain is smaller than that of the source domain;
And the prediction module is used for predicting the failure prediction value of the industrial equipment by utilizing the failure prediction model of the target domain, judging that the industrial equipment fails if the failure prediction value is larger than a preset threshold value, and judging that the industrial equipment is normal if the failure prediction value is not larger than the preset threshold value.
8. An industrial equipment failure prediction apparatus, comprising:
the second acquisition module is used for acquiring the operation data of the industrial equipment;
the input/output module is used for inputting the operation data into a trained fault prediction model and outputting a fault prediction value of the industrial equipment, wherein the fault prediction model is obtained by training the industrial equipment on the basis of a time sequence data set of a source domain and a target domain;
and the judging module is used for judging that the industrial equipment is faulty if the fault prediction value is larger than a preset threshold value, and judging that the industrial equipment is normal if the fault prediction value is not smaller than the preset threshold value.
9. An electronic device, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the industrial device fault prediction method of any one of claims 1-6.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the program is executed by a processor for implementing the industrial equipment failure prediction method according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211622060.8A CN116089870A (en) | 2022-12-16 | 2022-12-16 | Industrial equipment fault prediction method and device based on meta-learning under small sample condition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211622060.8A CN116089870A (en) | 2022-12-16 | 2022-12-16 | Industrial equipment fault prediction method and device based on meta-learning under small sample condition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116089870A true CN116089870A (en) | 2023-05-09 |
Family
ID=86198316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211622060.8A Pending CN116089870A (en) | 2022-12-16 | 2022-12-16 | Industrial equipment fault prediction method and device based on meta-learning under small sample condition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116089870A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116520814A (en) * | 2023-07-03 | 2023-08-01 | 清华大学 | Equipment fault prediction method and device based on federal learning under cloud edge cooperative architecture |
| CN116956197A (en) * | 2023-09-14 | 2023-10-27 | 山东理工昊明新能源有限公司 | Deep learning-based energy facility fault prediction method and device and electronic equipment |
| CN116881672B (en) * | 2023-09-05 | 2023-11-21 | 江西南昌济生制药有限责任公司 | Fault detection model training method and device, electronic equipment and storage medium |
| CN117114087A (en) * | 2023-10-23 | 2023-11-24 | 深圳开鸿数字产业发展有限公司 | Fault prediction method, computer device, and readable storage medium |
-
2022
- 2022-12-16 CN CN202211622060.8A patent/CN116089870A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116520814A (en) * | 2023-07-03 | 2023-08-01 | 清华大学 | Equipment fault prediction method and device based on federal learning under cloud edge cooperative architecture |
| CN116520814B (en) * | 2023-07-03 | 2023-09-05 | 清华大学 | Equipment fault prediction method and device based on federal learning under cloud edge cooperative architecture |
| CN116881672B (en) * | 2023-09-05 | 2023-11-21 | 江西南昌济生制药有限责任公司 | Fault detection model training method and device, electronic equipment and storage medium |
| CN116956197A (en) * | 2023-09-14 | 2023-10-27 | 山东理工昊明新能源有限公司 | Deep learning-based energy facility fault prediction method and device and electronic equipment |
| CN116956197B (en) * | 2023-09-14 | 2024-01-19 | 山东理工昊明新能源有限公司 | Deep learning-based energy facility fault prediction method and device and electronic equipment |
| CN117114087A (en) * | 2023-10-23 | 2023-11-24 | 深圳开鸿数字产业发展有限公司 | Fault prediction method, computer device, and readable storage medium |
| CN117114087B (en) * | 2023-10-23 | 2024-02-13 | 深圳开鸿数字产业发展有限公司 | Fault prediction method, computer device, and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116089870A (en) | Industrial equipment fault prediction method and device based on meta-learning under small sample condition | |
| Attarzadeh et al. | Proposing a new software cost estimation model based on artificial neural networks | |
| JP5005814B2 (en) | Battery long-term characteristic prediction system and method | |
| JP5313250B2 (en) | Battery long-term characteristic prediction system and method | |
| CN108509324B (en) | System and method for selecting computing platform | |
| JP2021047854A (en) | Method and apparatus for pruning neural network | |
| CN107977748B (en) | Multivariable distorted time sequence prediction method | |
| WO2022009010A1 (en) | Model fidelity monitoring and regeneration for manufacturing process decision support | |
| Zhao et al. | A neural architecture search method based on gradient descent for remaining useful life estimation | |
| CN112420125A (en) | Molecular attribute prediction method and device, intelligent equipment and terminal | |
| Zhou et al. | Aero-engine prognosis strategy based on multi-scale feature fusion and multi-task parallel learning | |
| CN113487019A (en) | Circuit fault diagnosis method and device, computer equipment and storage medium | |
| CN114787831B (en) | Improving accuracy of classification models | |
| CN107463486B (en) | System performance analysis method and device and server | |
| US10628541B2 (en) | Method and apparatus for selecting integrated circuit device neural network modeling sample | |
| CN116450486B (en) | Modeling method, device, equipment and medium for nodes in multi-element heterogeneous computing system | |
| CN112417463A (en) | Software vulnerability prediction method and device, computer equipment and storage medium | |
| WO2020055659A1 (en) | Generation and utilization of self-improving data-driven models with selective simulation of 3d object design | |
| CN113139332A (en) | Automatic model construction method, device and equipment | |
| CN111310907A (en) | Microwave assembly fault diagnosis method, device and equipment | |
| CN112925271A (en) | Data processing method, automatic control method and device | |
| CN113722975B (en) | Network model training method, room temperature prediction method, device, equipment and medium | |
| US11693921B2 (en) | Data preparation for artificial intelligence models | |
| CN114254764B (en) | Feedback-based machine learning model searching method, system, equipment and medium | |
| CN112861951B (en) | Image neural network parameter determining method and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |