Fault prediction method and device and computer readable storage medium

Abstract

The invention discloses a fault prediction method and device and a computer readable storage medium, and relates to the technical field of data processing, and the fault prediction method comprises the steps: obtaining the operation data of each to-be-tested device based on a sensor; determining features of the operation data by calculating quantiles of the operation data, the features including time features and event features; taking the features as input sequence data of a long short-term memory network, processing each input sequence data through the long short-term memory network according to the sequence of the input sequence data, and calculating the hidden state of each time step as an output sequence; and determining a predicted fault probability and / or a predicted fault period according to the output sequence. The technical effect of improving the accuracy of fault prediction is achieved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a fault prediction method, device, and computer-readable storage medium. Background Art

[0002] In equipment management, equipment failure prediction and cycle analysis are important links in equipment management. However, traditional failure prediction and cycle analysis methods mainly rely on the experience and judgment of experts. The prediction methods based on expert experience and judgment are often based on the personal experience and intuition of experts. Due to the lack of scientific data analysis and model support, traditional methods are difficult to accurately predict equipment failures and their occurrence cycles, which in turn affects the equipment maintenance plan and production schedule.

[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of this application is to provide a fault prediction method, device and computer-readable storage medium, aiming to solve the technical problem of how to improve the accuracy of fault prediction.

[0005] To achieve the above objectives, the present application proposes a fault prediction method, which includes: Acquire the operation data of each device under test based on sensors; Determine the characteristics of the operation data by calculating the quantiles of the operation data, wherein the characteristics include time characteristics and event characteristics; Using the features as input sequence data of a long short-term memory network, and processing each input sequence data through the long short-term memory network according to the order of the input sequence data, and calculating the hidden state of each time step as an output sequence; According to the output sequence, a predicted failure probability and / or a predicted failure period is determined.

[0006] In one embodiment, before the step of using the features as input sequence data of the long short-term memory network, and processing each input sequence data according to the order of the input sequence data through the long short-term memory network, and calculating the hidden state of each time step as the output sequence, the step includes: Based on at least one long short-term memory unit, construct an initial long short-term memory network, wherein the initial long short-term memory network unit includes a forget gate, an input gate, a candidate memory cell, and an output gate; Based on the historical operation data, fault records and maintenance information of the device under test, the initial long short-term memory network is trained to determine the long short-term memory network.

[0007] In one embodiment, the step of training the initial long short-term memory network based on the historical operation data, fault records and maintenance information of the device under test and determining the long short-term memory network includes: Preprocessing the historical operation data to determine a historical data sequence; According to the fault record and the maintenance information, define the label information of the historical data sequence and associate it with each corresponding historical data sequence; The initial long short-term memory network is trained by the historical data sequence, and the parameters of the network are updated by a back-propagation algorithm and an optimizer to determine the long short-term memory network.

[0008] In one embodiment, the step of training the initial long short-term memory network by the historical data sequence and updating the parameters of the network by a back propagation algorithm and an optimizer to determine the long short-term memory network includes: Calculating a loss function based on the historical data sequence to measure the difference between the prediction result of the initial long short-term memory network and the label information; Calculating the gradient of the loss function with respect to the parameter by the back propagation algorithm; According to the gradient of each of the parameters, the parameters are updated by the optimizer to minimize the loss function and determine the long short-term memory network.

[0009] In one embodiment, the step of acquiring the operating data of each device under test based on the sensor includes: Collecting the operating data of the device under test in real time through the sensor; Cleaning the running data, wherein the cleaning includes filling missing values ​​with interpolation, identifying and processing outliers, and deleting duplicate values; The operating data acquired by each sensor is converted into a unified format and merged into a unified data set.

[0010] In one embodiment, the step of determining the characteristics of the operating data by calculating the quantiles of the operating data, wherein the characteristics include time characteristics and event characteristics, comprises: Calculating the quantiles of the running data, and drawing a box plot according to the quantiles; Based on the box plot, the distribution of the operating data is determined and statistical features are extracted.

[0011] In one embodiment, the step of interpreting the box plot to determine the distribution of the operating data and extracting statistical features includes: Determine the dispersion of the operating data by using the interquartile range; or, Determine the center of the running data by the median; or, Assess the skewness of the running data by the length of the box and whiskers; or, Extreme values ​​of the operational data are identified based on the locations of the outliers.

[0012] In one embodiment, the fault prediction method further includes: If the predicted failure probability is output and the predicted failure probability exceeds a preset failure threshold, it is determined that the device under test fails, and a failure notification is sent to a management personnel; If the predicted failure period is output, the maintenance period and the inspection period of the device under test are determined according to the predicted failure period.

[0013] In addition, to achieve the above objectives, the present application also proposes a fault prediction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the fault prediction method described above.

[0014] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the fault prediction method described above are implemented.

[0015] The present application provides a fault prediction method, which first obtains the operation data of each device to be tested through a sensor, determines the time characteristics and event characteristics of the operation data by calculating the quantile of the operation data, uses the characteristics as the input sequence data of the long short-term memory network, and processes each input sequence data according to the order of the input sequence data through the long short-term memory network, calculates the hidden state of each time step as the output sequence, and then determines the predicted fault probability and / or predicted fault cycle. In the present application, the operation data of the device is collected in real time by the sensors deployed on the device, ensuring the reliability of the data source. By processing the collected operation data, calculating the quantile of the operation data, and further extracting the time characteristics and event characteristics of the data, it is helpful to convert the original data into useful information for fault prediction, reduce the data dimension, and improve the calculation efficiency. By inputting the extracted features into the long short-term memory network, the long short-term memory network processes the data of each time step one by one according to the order of the input sequence data, and calculates the hidden state of each time step. By utilizing the time series processing capability of the long short-term memory network, the changing trend of the operation state of the device over time is captured, and the potential laws and patterns in the operation state of the device are extracted. By further analyzing and processing the output sequence of the long short-term memory network to determine the predicted failure probability and / or predicted failure cycle of the equipment, the present application achieves the technical effect of improving the accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 A schematic diagram of a process flow provided for the first embodiment of the fault prediction method of the present application; Figure 2 A schematic diagram of a flow chart provided for the second embodiment of the fault prediction method of the present application; Figure 3 A schematic diagram of a process flow provided for the third embodiment of the fault prediction method of the present application; Figure 4 A schematic diagram of a process flow provided for the fourth embodiment of the fault prediction method of the present application; Figure 5 A schematic diagram of a box plot provided in Example 4 of the fault prediction method of the present application; Figure 6 A flowchart of the fifth embodiment of the fault prediction method of the present application is provided; Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the fault prediction method in the embodiment of the present application.

[0019] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0020] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0021] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0022] The main solutions of the embodiments of this application are: At present, traditional fault prediction and cycle analysis methods mainly rely on the experience and judgment of experts. The prediction methods based on expert experience and judgment are often based on the experts' personal experience and intuition. Due to the lack of scientific data analysis and model support, traditional methods are difficult to accurately predict equipment failures and their occurrence cycles, which in turn affects the equipment's maintenance plans and production arrangements.

[0023] This application uses a variety of sensors to obtain the operating data of each device under test in real time, ensuring the timeliness and accuracy of the data. By calculating the quantiles of the operating data, the distribution pattern and outliers of the data are revealed, thereby determining the characteristics of the data. The time characteristics reflect the changing trend of the data over time, and the event characteristics are related to the events or state changes of the device under test. The extracted features are used as the input sequence data of the long short-term memory network to capture the long-term dependencies in the time series data, thereby more accurately predicting the failure probability and failure cycle of the equipment, realizing the quantitative assessment of the equipment failure risk, and further improving the accuracy of fault prediction.

[0024] It should be noted that the execution subject of this embodiment may be a fault prediction device, or a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or a control device of a fault prediction device capable of realizing the above functions, etc., and this embodiment does not specifically limit this. The following takes the fault prediction device as the execution subject as an example to illustrate this embodiment and the following embodiments.

[0025] Embodiment 1 Based on this, the present application proposes a fault prediction method of the first embodiment, please refer to Figure 1 , the fault prediction method comprises: Step S10: acquiring operation data of each device to be tested based on the sensor.

[0026] The operating data of the equipment under test is collected in real time through sensors, the latest operating status of the equipment under test is captured, and a data source is provided for fault prediction.

[0027] In this embodiment, the sensor is a device for real-time monitoring of the operating status of the device under test, which can detect physical quantities and convert them into measurable signals, including physical quantities such as temperature, pressure, and vibration. The operating data is the raw data collected by the sensor that reflects the operating status of the device under test, including the measured values ​​of physical quantities such as temperature, pressure, vibration, and current.

[0028] For example, according to the type of the device under test and the need for fault prediction, select the appropriate sensor type, install the sensor in the key parts of the device under test, ensure that the operating status of the device under test is accurately captured, and configure the sampling frequency and data transmission method of the sensor. The sensor collects the operating data of the device under test in real time, including parameters such as temperature, pressure, vibration, and current, and records the operating time, maintenance records, and other information of the device under test.

[0029] Optionally, data cleaning and preprocessing are performed to ensure that the data input into subsequent analysis models is accurate, complete and consistent, thereby improving the accuracy and reliability of fault prediction of the equipment under test.

[0030] Furthermore, noise, outliers and missing values ​​in the data are removed through data cleaning and preprocessing, making the data more suitable for subsequent feature extraction and model training.

[0031] Exemplarily, the collected data is stored in a database, a preliminary check is performed, obvious errors and outliers are identified and removed, missing values ​​are filled or erroneous values ​​are corrected using statistical methods, duplicate data are removed, the data format is converted to ensure data consistency, and feature scaling is performed on numerical data to eliminate dimensional differences between different features.

[0032] Step S20, determining the characteristics of the operating data by calculating the quantiles of the operating data, wherein the characteristics include time characteristics and event characteristics.

[0033] Extracting key information that can reflect the operating status and failure mode of the equipment under test from a large amount of raw data, the feature sequence is the basis for subsequent model training and fault prediction, which simplifies the model complexity, improves the prediction efficiency, and enhances the generalization ability of the model. Through feature extraction, high-dimensional raw data is converted into low-dimensional feature vectors, which are easier to be understood and used by machine learning models.

[0034] In this embodiment, the quantile is the value at a specific position after a set of data is arranged from small to large, reflecting the distribution and concentration trend of the data. The time feature is the law and trend of data change over time, which is manifested as periodic fluctuations, trend growth or decline of data, etc. The event feature is the data feature related to a specific event or state change, which is manifested as sudden changes in data, abnormal peaks or valleys, etc.

[0035] As an optional implementation, the quartiles of the operating data are calculated to determine the overall distribution and dispersion of the data. The changing trend of the quantile values ​​over time is analyzed to determine the time characteristics of the operating data. Based on the quantile calculation results, outliers or mutation points in the data are identified, and these outliers or mutation points are combined with the operating records and maintenance records of the equipment under test to analyze their correlation with specific events or state changes to determine the event characteristics.

[0036] As another optional implementation, a linear regression method is used to identify long-term trends in the data, periodic components in the data are detected, and Fourier transform is used to identify periodic features, and mutation points or abnormal points in the data are identified, which represent significant changes in the state of the device under test or precursors to failure.

[0037] Step S30, using the features as input sequence data of the long short-term memory network, and processing each input sequence data through the long short-term memory network according to the order of the input sequence data, and calculating the hidden state of each time step as the output sequence.

[0038] By utilizing the powerful modeling capabilities of long short-term memory networks on time series data, the feature sequences can be further analyzed and predicted to obtain more accurate predictions of the operating status of the equipment under test, thereby timely discovering potential failure risks and improving the reliability of the equipment under test.

[0039] In this embodiment, the Long Short-Term Memory (LSTM) network is a special Recurrent Neural Network (RNN) architecture that overcomes the gradient vanishing or gradient exploding problems encountered by traditional RNNs when processing long time series by introducing a "gate" mechanism and cell states.

[0040] The LSTM network introduces three key gating mechanisms: forget gate, input gate, and output gate.

[0041] The forget gate decides what information to retain from the previous time step, reads the hidden state of the previous time step and the input of the current time step, and then generates a value between 0 and 1 through a sigmoid function, indicating the degree of retention of each state value.

[0042] The input gate determines which information to update in the current time step, and reads the hidden state of the previous time step and the input of the current time step to determine which new information will be stored. The input gate consists of two parts: a sigmoid layer, which determines which values ​​will be updated, and a tanh layer, which generates a new candidate value vector.

[0043] The output gate determines what the next hidden state output will be, reads the updated cell state and the input of the current time step, passes through a sigmoid layer to decide which cell states will be output, then passes through a tanh layer to generate candidate values ​​for the output state, and finally combines these two parts to form the final output.

[0044] As an optional implementation, the features extracted from the operating data of the device under test, including time features and event features, are organized into a sequence form as the input of the long short-term memory network. The long short-term memory network processes each feature vector step by step in chronological order. At each time step, the long short-term memory unit updates its internal state according to the current input feature vector, the hidden state and cell state of the previous time step, and generates the hidden state of the current time step, i.e., the output. After being processed by the long short-term memory network, an output sequence with the same length as the input sequence is obtained.

[0045] Step S40: determining a predicted failure probability and / or a predicted failure period according to the output sequence.

[0046] The prediction output of the long short-term memory network is converted into actual fault prediction information of the equipment under test. Through the long short-term memory network, the time series characteristics in the operation data of the equipment under test are captured, and then the probability of future failure of the equipment under test and the possible failure cycle are predicted, which is extremely valuable for formulating equipment maintenance plans, optimizing equipment operation strategies and reducing unexpected downtime.

[0047] In this embodiment, the predicted failure probability is a numerical representation of the possibility of equipment failure within a certain time period in the future predicted by the long short-term memory network based on historical operation data. The probability value is between 0 and 1, and the higher the value, the greater the possibility of failure. The predicted failure period is the approximate time interval before the next failure of the equipment under test is predicted by the long short-term memory network based on the operation mode of the equipment under test and historical failure data, which helps to formulate a preventive maintenance plan to ensure that the equipment under test is repaired in time before the failure occurs.

[0048] As an optional implementation, after obtaining the output sequence of the long short-term memory network, further processing is performed through one or more fully connected layers, and the final prediction result is calculated based on the output of the long short-term memory network to determine the probability of failure and / or failure cycle of the device under test.

[0049] Optionally, if the goal is to predict the probability of failure, the output layer is an activation function that maps the output to the interval (0, 1), representing the probability of failure occurring.

[0050] Optionally, if the goal is to predict the failure period, the output layer is a regression layer that directly outputs the predicted failure period value.

[0051] This embodiment provides a fault prediction method. This embodiment first obtains the operating data of each device under test in real time through multiple sensors to ensure the timeliness and accuracy of the data. By calculating the quantiles of the operating data, the distribution pattern and abnormal values ​​of the data are revealed, thereby determining the characteristics of the data. The time characteristics are used to reflect the change trend of the data over time. The event characteristics are related to the events or state changes of the device under test. The extracted features are used as the input sequence data of the long short-term memory network to capture the long-term dependencies in the time series data, thereby more accurately predicting the failure probability and failure cycle of the equipment, realizing the quantitative evaluation of the equipment failure risk, and further improving the accuracy of fault prediction.

[0052] Based on the first embodiment, the second embodiment of the present application proposes a fault prediction method, referring to Figure 2 , before step S30, including: Step S50, constructing an initial long short-term memory network based on at least one long short-term memory unit, wherein the initial long short-term memory network unit includes a forget gate, an input gate, a candidate memory cell and an output gate.

[0053] In this embodiment, the initial long short-term memory network is composed of multiple long short-term memory units, each of which has four core components built in: a forget gate, an input gate, a candidate memory cell, and an output gate. The gating mechanism enables the long short-term memory network to efficiently process time series data and capture long-term dependencies in the data.

[0054] For example, a detailed explanation of the components within the long short-term memory unit is as follows: The specific calculation formula of the forget gate is "f_t = \sigma(W_f \cdot [h_{t-1}, x_t] + b_f)", which receives the hidden state "h_{t-1}" of the previous moment and the input "x_t" of the current moment, performs linear transformation through the weight matrix "W_f" and the bias "b_f", and then outputs a vector "f_t" between 0 and 1 through the sigmoid activation function "\sigma". This vector determines how much information is forgotten from the previous memory cell state "C_{t-1}".

[0055] The specific calculation formula of the input gate is "i_t = \sigma(W_i \cdot [h_{t-1}, x_t] + b_i)", which receives the hidden state "h_{t-1}" of the previous moment and the input "x_t" of the current moment, performs linear transformation through the weight matrix "W_i" and bias "b_i", and then outputs a vector "i_t" through the sigmoid activation function "\sigma". This vector determines how much new information is written into the memory cell at the current moment.

[0056] The specific calculation formula for candidate memory cells is "\tilde{C}t = \tanh(W_C \cdot [h{t-1}, x_t]+ b_C)", which receives the hidden state "h_{t-1}" of the previous moment and the input "x_t" of the current moment, performs linear transformation through the weight matrix "W_C" and bias "b_C", and then processes it through the tanh activation function "\tanh" to generate a new candidate memory cell state "\tilde{C}t".

[0057] The specific calculation formula for updating memory cells is "C_t = f_t * C_{t-1} + i_t * \tilde{C}_t". According to the output of the forget gate "f_t" and the input gate "i_t", as well as the candidate memory cell state "\tilde{C}t", element-by-element multiplication "*" is used to combine this information and update the memory cell state "C_t".

[0058] The specific calculation formula of the output gate is "o_t = \sigma(W_o \cdot [h_{t-1}, x_t] + b_o)", which receives "h_{t-1}" and "x_t", performs linear transformation through the weight matrix "W_o" and the bias "b_o", and then outputs a vector "o_t" through the sigmoid activation function "\sigma". This vector controls how the memory cell state "C_t" affects the hidden state "h_t" at the current moment.

[0059] The specific calculation formula for the hidden state is "h_t = o_t * \tanh(C_t)". Based on the output gate "o_t" and the memory cell state "C_t", the hidden state "h_t" at the current moment is calculated through the tanh activation function "\tanh" and element-by-element multiplication "*".

[0060] As an optional implementation, at least one long short-term memory unit is stacked to form an initial long short-term memory network, the input of the network is the operating data sequence of the device under test, and the output of the network is the predicted failure probability or failure cycle of the device under test.

[0061] Optionally, the number of layers of the initial LSTM network, i.e., the number of stacked LSTM units, is adjusted according to the complexity of the task.

[0062] Step S60: training the initial long short-term memory network based on the historical operation data, fault records and maintenance information of the device under test, and determining the long short-term memory network.

[0063] Through training, the initial long short-term memory network can learn the patterns of equipment operating status changes over time, as well as the relationship between these patterns and equipment failures.

[0064] In this embodiment, historical operation data refers to various data generated by the operation of the device under test in the past period of time, including sensor readings such as temperature, pressure, vibration, and the operating time and load of the device under test, which reflect the operating status and performance of the device under test. Fault records record the fault events that occurred in the past of the device under test, including the time of the fault, the type of fault, the cause of the fault, etc. It is an important label data for training the LSTM network and is used to guide the network to learn the laws of fault occurrence. Maintenance information records the maintenance activities that the device under test has received in the past, including maintenance time, maintenance content, maintenance results, etc. Maintenance information helps the network understand the impact of maintenance activities on the operating status of the device under test, thereby more accurately predicting future faults.

[0065] Exemplarily, the historical operation data, fault records and maintenance information are cleaned and sorted, outliers and missing values ​​are removed, the data is converted into a format suitable for long short-term memory network input, and the data is normalized to improve the training efficiency of the model. According to the task requirements and data characteristics, the appropriate long short-term memory network structure is selected, the network's hidden unit number, learning rate, optimizer and other hyperparameters are set, and the preprocessed data is divided into training sets and test sets for training and verifying the model. The initial long short-term memory network is trained using the training set data, and the network parameters are optimized by the back propagation algorithm. During the training process, the model's loss function and accuracy and other indicators are monitored to evaluate the performance of the model. The trained model is verified using the test set data to evaluate its prediction performance. According to the verification results, the model is further tuned by adjusting hyperparameters, adding data enhancement and other tuning operations to determine the trained long short-term memory network.

[0066] Optionally, step S60 includes: Step S61, pre-processing the historical operation data to determine the historical data sequence.

[0067] Exemplarily, historical operating data of the equipment under test is obtained from the database. Records with a large number of missing values ​​are directly deleted. For numerical data, statistics such as mean, median, mode, etc. are used to fill missing values. Interpolation is used to estimate missing values ​​of time series data. Convert the data into a form suitable for analysis, convert text types into numerical types, convert date and time types into timestamps, and standardize numerical data so that the data are on the same scale. Arrange the data in chronological order according to the timestamp or date field, select specific sensor readings, performance indicators, etc. as components of the data series according to business needs, and aggregate the data according to time intervals to form a more concise data series.

[0068] Step S62: define label information of the historical data sequence according to the fault record and the maintenance information, and associate it with each corresponding historical data sequence.

[0069] By adding labels to historical data sequences, the operating status of the equipment under test is associated with known faults or maintenance events, so that the model can learn the mapping relationship from operating status to faults or maintenance requirements. In this embodiment, the label information is information added to the historical data sequence, indicating the fault or maintenance requirement corresponding to the data sequence. The label information includes classification labels such as fault type and maintenance requirement and regression labels such as probability of fault occurrence and remaining service life.

[0070] For example, the timestamps in the fault records and maintenance information are aligned with the timestamps in the historical data series. According to business needs and analysis objectives, appropriate labels are defined, with the fault type as the classification label and the remaining service life as the regression label. According to the timestamps and content in the fault records and maintenance information, corresponding labels are assigned to each historical data series for association. Classification labels are assigned according to fault types or maintenance requirements, and regression labels are defined based on the probability of fault occurrence, predicted value of remaining service life, and other logic.

[0071] Optionally, for information that has no direct timestamp association, such as event-based records, an association with a historical data sequence is established through the ID of the device under test, a serial number, and the like.

[0072] Optionally, the historical data sequence with label information is split into training set, validation set and test set.

[0073] Step S63, training the initial long short-term memory network through the historical data sequence, and updating the parameters of the network through a back propagation algorithm and an optimizer to determine the long short-term memory network.

[0074] Exemplarily, a historical data sequence with label information is input into an initial long short-term memory network for training, and the gradient is calculated by a back-propagation algorithm based on the error between the output result and the label, and the parameters of the network are updated according to the gradient by an optimizer.

[0075] Optionally, step S63 includes: Step A11, calculating a loss function based on the historical data sequence to measure the difference between the prediction result of the initial long short-term memory network and the label.

[0076] In this embodiment, the loss function is a function that measures the difference between the initial LSTM network prediction result and the true label, including a cross entropy loss function for classification tasks and a mean square error loss function for regression tasks.

[0077] As an optional implementation, the historical data sequence is used as input, forward propagation is performed through the initial long short-term memory network, the network's prediction result is obtained, and the difference between the prediction result and the true label, that is, the value of the loss function, is calculated.

[0078] Exemplarily, for a classification task, the output layer of the network is set to have a number of neurons equal to the number of categories, and the output is converted into a probability distribution using a softmax function, and the cross entropy loss function is calculated.

[0079] Exemplarily, for a regression task, the output layer of the network is set to one neuron or multiple neurons, and the predicted value is directly output to calculate the mean square error loss function.

[0080] Step A12, calculating the gradient of the loss function with respect to the parameter by a back propagation algorithm.

[0081] In this embodiment, the back propagation algorithm is an algorithm for calculating the gradient of the loss function with respect to the network parameters. Through back propagation, the error of the loss function is back propagated from the output layer to the input layer, thereby calculating the gradient of each parameter. The gradient is the partial derivative of the loss function with respect to a certain parameter, which represents the rate of change of the loss function in the direction of the parameter.

[0082] Optionally, before calculating the gradient of the loss function with respect to the parameters, the value of the loss function is monitored. If the value of the loss function is large, it means that the difference between the prediction result of the network and the true label is large, and the network parameters need to be updated.

[0083] Exemplarily, a back-propagation algorithm is used, starting from the output layer, to calculate the gradient of the loss function with respect to each parameter layer by layer.

[0084] Step A13, according to the gradient of each parameter, updating the parameter through the optimizer to minimize the loss function and determine the long short-term memory network.

[0085] In this embodiment, the optimizer is an RMSprop (Root Mean Square Propagation) optimizer, which is an adaptive learning rate optimization algorithm for updating network parameters. The parameters are updated according to the learning rate based on the gradient of each parameter to minimize the loss function.

[0086] Exemplarily, for each parameter, the root mean square (RMS) of its gradient is calculated, the RMS value is used to adjust the learning rate of each parameter, and the calculated adaptive learning rate and gradient are used to update the weights and bias items of the network.

[0087] As an example of this embodiment, before the data is input into the initial long short-term memory network, a series of preprocessing steps such as data cleaning, missing value processing, normalization or standardization are performed, and the preprocessed data is divided into a training set, a validation set and a test set. The training set is used to train the model, the validation set is used to monitor the performance of the model and adjust the hyperparameters during the training process, and the test set is used to evaluate its generalization ability after the model training is completed. Constructing the initial long short-term memory network structure includes determining the configuration of the input layer, hidden layer and output layer, and selecting the activation function, etc. The training data is input into the initial long short-term memory network, and the output is obtained by layer-by-layer calculation. The loss function value is calculated according to the output of the network and the true label. According to the loss function value, the gradient of the network parameters is calculated by the chain rule, and the network parameters are updated according to the gradient using the optimizer to minimize the loss function value. During the training process, the change of the loss function value is recorded regularly, and the performance of the initial long short-term memory network model at the end of each training cycle is evaluated using the validation set, and the hyperparameters or optimizer settings are adjusted accordingly. After the long short-term memory network model training is completed, the test set is used for the final evaluation.

[0088] This embodiment provides a fault prediction method. This embodiment first constructs a long short-term memory network unit including a forget gate, an input gate, a candidate memory cell, and an output gate to effectively capture the time series characteristics and long-term dependencies in the historical operation data. By cleaning and preprocessing the historical operation data, the data quality is improved, and the error in the initial long short-term memory network model training process is reduced, thereby enhancing the accuracy of the model. The historical data sequence is labeled according to the fault record and maintenance information, so that the model can learn the data characteristics before and after the fault occurs. The difference between the network prediction result and the label is measured by calculating the loss function, and the performance of the model is further quantified. The gradient of the loss function with respect to the network parameters is calculated by the back propagation algorithm, and the parameters are updated according to the gradient, so that the loss function value can be gradually reduced, thereby improving the prediction accuracy of the model. The optimizer is used to automatically update the model parameters according to the gradient to improve the efficiency and accuracy of model training.

[0089] Based on the first embodiment, the third embodiment of the present application proposes a fault prediction method, referring to Figure 3 , step S10 comprises: Step S11, collecting the operating data of the device under test in real time through the sensor.

[0090] Real-time monitoring of the operating data of the equipment under test is an important basis for evaluating its performance, predicting potential failures, and optimizing maintenance strategies.

[0091] In this embodiment, the operating data refers to various data generated by the device under test during operation, including but not limited to the measured values ​​of physical quantities such as temperature, pressure, flow, vibration, current, voltage, and the operating status and working time of the device under test.

[0092] Exemplarily, according to the type of the device to be tested and the physical quantity to be monitored, select a suitable sensor, install the sensor on the device to be tested, and ensure that the sensor can accurately measure the required physical quantity. Set up a data acquisition system to receive and process the data transmitted by the sensor. Connect the sensor to the data acquisition system to ensure that the sensor can transmit data to the data acquisition system in real time. According to the monitoring requirements, set the parameters of the data acquisition system such as sampling frequency, data format, and data storage method, and collect the operating data of the device to be tested in real time according to the set parameters.

[0093] Optionally, sensors are selected. For mechanical equipment, vibration sensors, temperature sensors, pressure sensors, etc. are selected; for electrical equipment, current sensors, voltage sensors, etc. are selected.

[0094] Optionally, the data acquisition system can be a dedicated hardware device, or a software program such as a data acquisition card, data acquisition software, etc.

[0095] Optionally, during the process of collecting the operating data of the device under test, the status of the sensor and the operating status of the data acquisition system are regularly checked to ensure the accuracy and completeness of the data.

[0096] Step S12, cleaning the operating data, wherein the cleaning includes filling missing values ​​with interpolation, identifying and processing outliers, and deleting duplicate values.

[0097] Comprehensive data cleaning of the operating data of the equipment under test helps to eliminate noise, missing values, outliers, duplicate values ​​and other problems in the data, thereby ensuring data quality and providing a solid foundation for subsequent data analysis, fault diagnosis, performance evaluation and predictive model training.

[0098] In this embodiment, interpolation is a method for estimating and filling missing values ​​in a data set. By using the relationship between known data points, reasonable estimates of missing data points are calculated. Missing values ​​refer to unknown or missing values ​​of certain data items in a data set. Outliers are data points in a data set that appear abnormal or illogical compared to most data points due to sensor failure or data recording errors. Duplicate values ​​refer to data items in a data set that have values ​​that are exactly the same or highly similar. These data items are considered duplicate data.

[0099] Exemplarily, according to the characteristics of the data and the distribution of missing values, interpolation filling is performed using mean, median, and mode filling methods, outliers in the data set are identified by setting thresholds or statistical methods, and outliers are deleted, replaced, or re-recorded, and duplicate records are found and deleted.

[0100] Step S13: converting the operation data acquired by each sensor into a unified format and merging them into a unified data set.

[0101] In the process of data collection and processing, operational data generated from different sources, at different time points or in different systems may use different formats, units or data structures. This inconsistency will bring great difficulties to data analysis and may even lead to erroneous analysis results. Converting these data into a unified format and merging them into one data set can ensure the coherence, consistency and comparability of the data, thereby simplifying the data analysis process and improving the accuracy and efficiency of the analysis.

[0102] During the merging process, mapping relationships between data sources are established to solve problems such as homonyms and synonyms, and ensure the accuracy and consistency of data.

[0103] In this embodiment, the operating data comes from multiple sensors, databases, etc. from multiple different data sources, and these data sources need to be merged into a unified data set.

[0104] Exemplarily, each data source is analyzed in detail to understand the name, meaning, data type, unit and other information of the data item. Based on the analysis results, a mapping relationship between data sources is established to solve the problem of homonymy and heteronymy. Based on the mapping relationship between data sources, the data from different data sources are merged into a unified data set, and the original meaning and unit of the data item are maintained. Redundant data in the data set is detected using methods such as correlation analysis. For the detected redundant data, further review and analysis are performed to delete the repeated attributes or data items of the redundant data.

[0105] This embodiment provides a fault prediction method. This embodiment first ensures the timeliness and comprehensiveness of the data by collecting operation data in real time, fills in missing values ​​through interpolation, effectively reduces gaps in the data, and improves data integrity. It identifies and processes abnormal values, removes unreasonable data caused by sensor failures or data recording errors, and enhances data accuracy. It avoids data redundancy by deleting duplicate values. It converts operation data from different sources into a unified format and merges them into a unified data set, eliminating format differences between data and improving data consistency, which provides a basis for subsequent data analysis and processing.

[0106] Based on the first embodiment, the fourth embodiment of the present application proposes a fault prediction method, referring to Figure 4 , step S40 comprises: Step S41, calculating the quantiles of the operating data, and drawing a box plot according to the quantiles.

[0107] Draw a box plot based on the quantiles of the running data to intuitively display the distribution characteristics of the data, including the central trend of the data, the degree of dispersion, and the presence of outliers.

[0108] In this embodiment, quantile refers to the numerical point that divides the probability distribution range of a random variable into several equal parts, that is, the numerical value at a certain percentile after the data is sorted from small to large. Quartiles include the 25th percentile, the 50th percentile, i.e. the median, and the 75th percentile, which divide the data set into four equal-frequency parts. The box plot is used to display a statistical chart of a set of data dispersion information, consisting of five numerical points, the minimum value (Min), the first quartile (Q1), the median (Median, M), the third quartile (Q3) and the maximum value (Max). The box plot intuitively shows the center position (median) of the data, the degree of dispersion of the data (interquartile range), and outliers (values ​​beyond the range of "Q1-1.5IQR" or "Q3+1.5IQR", IQR is the interquartile range, i.e. "Q3-Q1").

[0109] For example, please refer to Figure 5 , calculate the key quartiles such as the first quartile (Q1), median (Q2), and third quartile (Q3) of the preprocessed running data, and calculate the interquartile range (IQR) based on the difference between the third quartile (Q3) and the first quartile (Q1), that is, "IQR = Q3 - Q1". Draw the upper and lower boundaries of the box, with the upper boundary being the third quartile (Q3) and the lower boundary being the first quartile (Q1). Within the box, draw a horizontal line from the lower boundary to the upper boundary, namely the median line, indicating the median (M). Starting from the upper and lower boundaries of the box, draw two line segments, namely the box whiskers, respectively, to indicate the overall distribution range of the data.

[0110] Step S42, interpreting the box plot to determine the distribution of the operating data and extracting statistical features.

[0111] By interpreting the box plot, we can deeply analyze the distribution of operating data and extract key statistical features to understand the inherent laws and potential problems of the data.

[0112] Optionally, step S42 includes: Step A21, determining the degree of dispersion of the operating data by using the interquartile range.

[0113] In this embodiment, the interquartile range reflects the variation range of data points in the data set and is an important indicator for measuring the degree of data dispersion.

[0114] For example, according to the box plot analysis of the interquartile range, the larger the interquartile range, the higher the degree of dispersion of the data; the smaller the interquartile range, the lower the degree of dispersion of the data.

[0115] Step A22: determining the data center of the operating data by using the median.

[0116] In this embodiment, the median is the middle number or the average of two middle numbers after the data set is sorted from small to large, and is a robust central tendency measure that is not affected by extreme values.

[0117] Exemplarily, the median value of the box plot is analyzed to determine the central tendency of the operating data.

[0118] Step A23, evaluating the skewness of the running data by the length of the box whisker.

[0119] In this embodiment, the box-whisker length is the length of the box whiskers extending from the upper and lower boundaries of the box in the box plot, indicating the overall distribution range of the data and reflecting the degree of skewness of the data.

[0120] For example, based on the length of the box whiskers in the box plot, if one side of the box whiskers is significantly longer than the other side, it indicates that the data is skewed, and the direction and degree of the skewness are further analyzed to determine the distribution characteristics of the data.

[0121] Step A24, identifying the extreme values ​​of the operating data according to the positions of the abnormal values.

[0122] Exemplarily, data points outside the box and whiskers in the box plot are identified as outliers, and the locations and numbers of outliers are recorded to further identify extreme cases in the data.

[0123] This embodiment provides a fault prediction method. This embodiment first calculates quantiles and draws a box plot to achieve data visualization. The box plot can accurately display the central trend, dispersion, skewness and outliers of the data. The extreme values ​​or outliers in the data can be intuitively identified through the length of the box whisker and the mark of the outlier. Based on the information of the box plot, a series of key statistical features can be extracted to provide a basis for subsequent data analysis or model building.

[0124] Based on the first embodiment, the fifth embodiment of the present application proposes a fault prediction method, referring to Figure 6 , after step S40, comprising: Step S70: If the predicted failure probability is output and the predicted failure probability exceeds a preset failure threshold, it is determined that the device under test fails, and a failure notification is sent to a management personnel.

[0125] In this embodiment, the preset fault threshold is a threshold determined based on factors such as reliability requirements, safety standards, maintenance costs, and downtime of the device or system for comparison with the predicted fault probability. Fault notification is a notification mechanism for reporting to management personnel information that the device under test has or is about to fail. The notification method includes email, text message, phone call, internal system message, etc. The notification content includes information such as the name, location, predicted fault type, and predicted fault time of the device.

[0126] Exemplarily, the output predicted failure probability is compared with a preset failure threshold. When the predicted failure probability exceeds the failure threshold, it is determined that the device is about to fail or is already in a failure state, and a failure notification is sent to the management personnel. If the predicted failure probability is less than or equal to the failure threshold, the operating data of the device under test is continuously monitored and the predicted failure probability is updated.

[0127] Step S80: If the predicted failure period is output, determine the maintenance period and the inspection period of the device under test according to the predicted failure period.

[0128] In this embodiment, the maintenance cycle is a period of preventive maintenance activities such as regular inspection, cleaning, lubrication, etc. of the equipment. The overhaul cycle is a period of more comprehensive inspection, repair or replacement of worn parts of the equipment.

[0129] Exemplarily, the maintenance cycle and the inspection cycle are set based on the output predicted failure cycle and factors such as the type, purpose, and operating environment of the device under test.

[0130] This embodiment provides a fault prediction method, which first compares the predicted fault probability with the fault threshold to timely discover the potential fault risk of the device under test and provide timely warning information to the management personnel. By setting the maintenance cycle and the inspection cycle based on the predicted fault cycle, the preventive maintenance of the device under test can be achieved, and the faults caused by aging or wear of the device under test can be reduced, thereby improving the reliability of the device under test.

[0131] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the fault prediction method of the present application. More simple transformations based on this technical concept are all within the protection scope of the present application.

[0132] The present application provides a fault prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the fault prediction method in the above-mentioned embodiment one.

[0133] Reference below Figure 7 , which shows a schematic diagram of the structure of a fault prediction device suitable for implementing the embodiment of the present application. The fault prediction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The fault prediction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0134] like Figure 7 As shown, the fault prediction device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the fault prediction device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the fault prediction device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a fault prediction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0135] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0136] The fault prediction device provided by the present application adopts the fault prediction method in the above embodiment, which can solve the technical problem of how to improve the accuracy of fault prediction. Compared with the prior art, the beneficial effects of the fault prediction device provided by the present application are the same as the beneficial effects of the fault prediction method provided by the above embodiment, and the other technical features in the fault prediction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0137] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0138] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0139] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the fault prediction method in the above-mentioned embodiment.

[0140] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0141] The computer-readable storage medium may be included in the fault prediction device; or may exist independently without being assembled into the fault prediction device.

[0142] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the fault prediction device, the fault prediction device can write computer program codes for performing the operations of the present application in one or more programming languages ​​or a combination thereof. The programming languages ​​include object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0143] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0144] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0145] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned fault prediction method, and can solve the technical problem of how to improve the accuracy of fault prediction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the fault prediction method provided in the above-mentioned embodiment, and will not be described in detail here.

[0146] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A fault prediction method, characterized in that: The fault prediction method comprises: Acquire the operation data of each device under test based on sensors; Determine the characteristics of the operation data by calculating the quantiles of the operation data, wherein the characteristics include time characteristics and event characteristics; Using the features as input sequence data of a long short-term memory network, and processing each input sequence data through the long short-term memory network according to the order of the input sequence data, and calculating the hidden state of each time step as an output sequence; According to the output sequence, a predicted failure probability and / or a predicted failure period is determined.

2. The fault prediction method according to claim 1, characterized in that: Before the step of using the features as input sequence data of the long short-term memory network, and processing each input sequence data according to the order of the input sequence data through the long short-term memory network, and calculating the hidden state of each time step as the output sequence, the step includes: Based on at least one long short-term memory unit, construct an initial long short-term memory network, wherein the initial long short-term memory network unit includes a forget gate, an input gate, a candidate memory cell and an output gate; Based on the historical operation data, fault records and maintenance information of the device under test, the initial long short-term memory network is trained to determine the long short-term memory network.

3. The fault prediction method according to claim 2, characterized in that: The step of training the initial long short-term memory network based on the historical operation data, fault records and maintenance information of the device under test and determining the long short-term memory network comprises: Preprocessing the historical operation data to determine a historical data sequence; According to the fault record and the maintenance information, define the label information of the historical data sequence and associate it with each corresponding historical data sequence; The initial long short-term memory network is trained by the historical data sequence, and the parameters of the network are updated by a back-propagation algorithm and an optimizer to determine the long short-term memory network.

4. The fault prediction method according to claim 3, characterized in that: The steps of training the initial long short-term memory network by the historical data sequence and updating the parameters of the network by the back propagation algorithm and the optimizer to determine the long short-term memory network include: Calculating a loss function based on the historical data sequence to measure the difference between the prediction result of the initial long short-term memory network and the label information; Calculating the gradient of the loss function with respect to the parameter by the back propagation algorithm; According to the gradient of each of the parameters, the parameters are updated by the optimizer to minimize the loss function and determine the long short-term memory network.

5. The fault prediction method according to claim 1, characterized in that: The step of obtaining the operating data of each device under test based on the sensor includes: Collecting the operating data of the device under test in real time through the sensor; Cleaning the running data, wherein the cleaning includes filling missing values ​​with interpolation, identifying and processing outliers, and deleting duplicate values; The operating data acquired by each sensor is converted into a unified format and merged into a unified data set.

6. The fault prediction method according to claim 1, characterized in that: The step of determining the characteristics of the operating data by calculating the quantiles of the operating data, wherein the characteristics include time characteristics and event characteristics, comprises: Calculating the quantiles of the running data, and drawing a box plot according to the quantiles; Based on the box plot, the distribution of the operating data is determined and statistical features are extracted.

7. The fault prediction method according to claim 6, characterized in that: The step of interpreting the box plot to determine the distribution of the operating data and extracting statistical features includes: Determine the dispersion of the operating data by using the interquartile range; or, Determine the center of the running data by the median; or, Assess the skewness of the running data by the length of the box and whiskers; or, The extreme values ​​of the operating data are identified based on the locations of the outliers.

8. The fault prediction method according to claim 1, characterized in that: The fault prediction method further includes: If the predicted failure probability is output and the predicted failure probability exceeds a preset failure threshold, it is determined that the device under test fails, and a failure notification is sent to a management personnel; If the predicted failure period is output, the maintenance period and the inspection period of the device under test are determined according to the predicted failure period.

9. A fault prediction device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the fault prediction method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the fault prediction method according to any one of claims 1 to 8 are implemented.

Images