Fault prediction method, computer storage medium and terminal device

Abstract

The present invention relates to a fault prediction method, a computer storage medium and a terminal device, and constructs and trains a prediction model with fluctuation parameters, stable parameters and statistical indicators of each fluctuation parameter as input and normal and fault probabilities of equipment as output, and then collects the current operation data of the equipment to input the prediction model to obtain a prediction result. The parameters involve uncontrollable fluctuation parameters and controllable stable parameters, and are combined with the statistical indicators of the fluctuation parameters to comprehensively reflect the internal connections between the various parameters of the equipment and extract the intricate internal features. It is a fault predictive maintenance with high prediction performance and high accuracy, and adopts advanced data analysis technology and machine learning algorithms. By monitoring the operating status, historical data and other relevant factors of the equipment, it predicts possible future failures or performance degradation of the equipment, thereby taking preventive maintenance measures before the failure occurs. A maintenance strategy; specifically forward-looking and feasible.

Description

Technical Field

[0001] The present invention relates to the field of fault prediction, and in particular to a fault prediction method, a computer storage medium and a terminal device. Background Art

[0002] With the development of automation and intelligence in industrial production, equipment failures have an increasingly greater impact on production operations. If equipment failures occur frequently, they will lead to production line shutdowns, reduced production efficiency, and other problems, increasing maintenance costs and production losses. Traditional equipment maintenance mainly relies on regular inspections and emergency repairs after failures occur, lacking effective preventive maintenance measures. This regular inspection and maintenance method can no longer meet the needs of equipment maintenance, and equipment failure information cannot be obtained and analyzed in a timely manner, resulting in difficulties in fault prediction and early warning, and inability to take maintenance measures in advance.

[0003] Therefore, how to implement predictive maintenance methods, realize the prediction and early warning of equipment failures, discover potential failure risks in advance, reduce equipment failure rates, and reduce maintenance costs and production losses is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a fault prediction method, comprising:

[0005] Collect historical operation data of the equipment under normal and fault conditions, including fluctuation parameters, stable parameters and statistical indicators of each fluctuation parameter, and mark them as normal data or fault data to build a training data set;

[0006] Construct a prediction model with fluctuation parameters, stability parameters and statistical indicators of each fluctuation parameter as input and equipment normal and failure probabilities as output; and obtain a trained prediction model based on the training data set;

[0007] Collect the current operating data of the equipment, including fluctuation parameters and stability parameters; and calculate the statistical indicators of each fluctuation parameter;

[0008] The current fluctuation parameters, stability parameters and statistical indicators of each fluctuation parameter are input into the trained prediction model, and the normal and failure probabilities of the equipment are output to obtain the fault prediction results.

[0009] Further, the prediction model includes: an input module, a learning vector quantization module, a Bayesian module, a deep learning neural network module and an output module;

[0010] The input ends of the learning vector quantization module, the Bayesian module and the deep learning neural network module are connected to the input module, and the output ends are connected to the input end of the output module; the output probabilities of the learning vector quantization module, the Bayesian module and the deep learning neural network module are combined to output the probability of normal and faulty equipment.

[0011] Furthermore, the input module includes:

[0012] A first input unit connected to an input terminal of the learning vector quantization module to input a stable parameter into the learning vector quantization module;

[0013] A second input unit is connected to an input terminal of the Bayesian module to input the statistical index into the Bayesian module;

[0014] The third input unit is connected to the input end of the deep learning neural network module to input the fluctuation parameter, the stability parameter and the statistical indicators of each fluctuation parameter into the deep learning neural network module.

[0015] Furthermore, in the output module, the learning vector quantization module, the Bayesian module and the deep learning neural network module are configured with their own fully connected layers, and then the superposition module is linked, and the total probability obtained by superposition is the final output result of the prediction model;

[0016] Or, in the output module, first superimpose the outputs of the learning vector quantization module and the deep learning neural network module, then link the fully connected layer to obtain the total output of the vector quantization module and the deep learning neural network module, then link the superposition module, superimpose the output of the Bayesian module, and superimpose the total probability obtained as the final output result of the prediction model.

[0017] Furthermore, the deep neural network module includes three layers of convolutional units connected in sequence and a pooling layer connected at the end; each convolutional unit includes a layer of convolution kernel and a binary classification activation function.

[0018] Furthermore, for the first convolution unit, the convolution kernel is 3×3, the step size is 3, and the padding is 2; for the second convolution unit, the convolution kernel is 5×5, the step size is 4, and the padding is 2; for the third convolution unit, the convolution kernel is 7×7, the step size is 5, and the padding is 2; finally, after the fully connected layer, the first probability y1(q1,q2) is output; q1 is the first normal probability, and q2 is the first fault probability.

[0019] Furthermore, when the failure probability is greater than or equal to the failure threshold, it means that the device prediction result is a failure; otherwise, it means that the device prediction result is normal.

[0020] Further, the fluctuating parameters include any one or more of temperature, pressure, vibration, current, and voltage;

[0021] Stability parameters, including any one or more of equipment service life, equipment historical failure records, equipment maintenance records, equipment production years, equipment model, and equipment configuration;

[0022] Statistical indicators include any one or more of the difference, amplitude, and phase of each fluctuation parameter.

[0023] On the other hand, the present invention further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned fault prediction methods.

[0024] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores program codes that can be executed by the processor; the program codes are used to execute any of the above-mentioned fault prediction methods.

[0025] The fault prediction method, computer storage medium and terminal device of the present invention are based on the historical operation data of the equipment under normal and faulty states, including fluctuation parameters, stable parameters and statistical indicators, and construct and train a prediction model with fluctuation parameters, stable parameters and statistical indicators of each fluctuation parameter as input and normal and faulty probability of the equipment as output, and then collect the current operation data of the equipment to input the prediction model to obtain the prediction result, whose parameters involve uncontrollable fluctuation parameters and controllable stable parameters, and combined with the statistical indicators of fluctuation parameters to comprehensively reflect the internal connection between the various parameters of the equipment and extract the intricate internal characteristics. It is a fault predictive maintenance with high prediction performance and high accuracy, which adopts advanced data analysis technology and machine learning algorithms, and predicts possible future failures or performance degradation of the equipment by monitoring the operating status, historical data and other relevant factors of the equipment, so as to take preventive maintenance measures before the failure occurs. A maintenance strategy; specific foresight and feasibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of an embodiment of a fault prediction method of the present invention;

[0027] Figure 2 A schematic diagram of the structure of an embodiment of the prediction model of the present invention;

[0028] Figure 3 A schematic diagram of the structure of an embodiment of an input module of a prediction model of the present invention;

[0029] Figure 4 It is a structural schematic diagram of another embodiment of the prediction model of the present invention;

[0030] Figure 5 A schematic diagram of the structure of an embodiment of a deep learning neural network module of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] It should be noted that if the embodiments of the present invention involve directional indications, such as up, down, left, right, front, back, etc., then the directional indication is only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. In addition, if the embodiments of the present invention involve descriptions of "first, second", "S1, S2", "step one, step two", etc., then such descriptions are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of indicated technical features or indicating the execution order of the method, etc. Those skilled in the art can understand that anything that does not violate the gist of the invention under the technical concept of the invention should be included in the protection scope of the present invention.

[0033] like Figure 1 As shown, the present invention provides a fault prediction method, comprising:

[0034] S1: Collect historical operation data of the equipment under normal and fault conditions, including fluctuation parameters, stable parameters and statistical indicators of each fluctuation parameter, and mark them as normal data or fault data to build a training data set;

[0035] Preferably, historical operation data of the equipment is collected when the equipment is in a normal state or in a fault state, and the normal data label may be 0 and the fault data label may be 1 for marking.

[0036] More preferably, the fluctuation parameters may include any one or more of temperature, pressure, vibration, current, and voltage, which are fluctuation data that are unstable over time; the stable parameters may include any one or more of equipment service life, equipment historical failure records, equipment maintenance records, equipment production years, equipment models, and equipment configurations, which are basic data that are stable over time. The statistical indicators may include, but are not limited to, any one or more of the difference, amplitude, and phase of each fluctuation parameter.

[0037] Specific: The calculation formula of statistical indicators is as follows:

[0038] 1. Delta: It indicates the difference between two adjacent data points and is often used to detect signal changes and anomalies. The calculation formula for the delta is:

[0039] Δ = x(t) - x(t-1)

[0040] Where Δ represents the difference, x(t) represents the value of the t-th data point, and x(t-1) represents the value of the t-1-th data point.

[0041] 2. Amplitude: Indicates the height of the waveform, often used to describe the strength and energy of the signal. The calculation formula for amplitude is:

[0042] A = max(x(t)) - min(x(t))

[0043] Where A represents the amplitude, max(x(t)) represents the maximum value of x(t), and min(x(t)) represents the minimum value of x(t).

[0044] 3. Phase: It indicates the time offset of the waveform and is often used to describe the relative position and timing relationship of the signal. The calculation formula of phase is:

[0045] Φ = ∠(x(t)) where Φ represents the phase and ∠(x(t)) represents the argument of x(t), that is, the angle between x(t) and the positive real axis.

[0046] More preferably, after data collection, data cleaning and preprocessing are also included; examples include removal of outliers, missing value processing, data smoothing, etc., to ensure the accuracy and completeness of the data.

[0047] In a specific embodiment, taking all the parameters in the above example as an example, the data includes fluctuation parameters: temperature, pressure, vibration, current, voltage, a total of 5, represented by a1-a5, stable parameters: equipment service life, equipment historical failure record, equipment maintenance record, equipment production year, equipment model, equipment configuration, a total of 6, represented by a6-a 11 Indicates; Statistical indicators: the difference, amplitude, and phase of temperature, pressure, vibration, current, and voltage within a period of time, a total of 15, with a 12 -a 26 express.

[0048] More preferably, the data may be selected as The feature matrix X is represented by N, where N represents the number of parameters and M represents time, such as days or hours. Taking the above 26 variables as an example, a 26×12 feature matrix can be selected to represent the 26 variables collected for 12 consecutive days.

[0049] S2: Construct a prediction model with fluctuation parameters, stability parameters and statistical indicators of each fluctuation parameter as input and equipment normal and failure probabilities as output; and obtain the trained prediction model based on the training data set;

[0050] A: Steps to build a prediction model:

[0051] Specifically, the prediction model can use any data analysis technology and machine learning algorithm. Figure 2 As shown, the present invention provides a prediction model combining three forms, including: an input module, a learning vector quantization module LVQ, a Bayesian module NBC, a deep learning neural network module and an output module;

[0052] The input ends of the learning vector quantization module, the Bayesian module and the deep learning neural network module are connected to the input module, and the output ends are connected to the input end of the output module; the output probabilities of the learning vector quantization module, the Bayesian module and the deep learning neural network module are combined to output the probability of normal and faulty equipment.

[0053] In this embodiment, a preferred embodiment of the prediction model is given. Although learning vector quantization LVQ, Bayesian NBC and deep neural network are commonly used algorithms for data analysis and machine learning, the present invention considers the specific structural form of the specific content of the sampled data, involving fluctuation parameters, stability parameters and statistical indicators, and selects the above three modules to form a prediction module to adapt to the actual application scenario of fault prediction of the present invention: vector quantization is selected to map the points in the data set to a discrete point set, that is, a codebook; Bayesian statistical method is selected; deep neural network is selected to learn the complex connections between data; it is not a conventional technical selection, but starting from the actual characteristics of the input content, focusing on fluctuation parameters, stability parameters and statistical parameters respectively, comprehensively extracting the unique characteristics of fluctuation parameters, stability parameters and statistical indicators, and then comprehensively considering the outputs of the three modules combined as the total output of the prediction model, which can further improve the prediction accuracy of the prediction model.

[0054] (1) Input module:

[0055] More preferably, in the input module, the three types of data - fluctuation parameters, stability parameters and statistical parameters - can be directly input into the learning vector quantization module, the Bayesian module and the deep learning neural network module, or can be input in a targeted manner. Figure 3 As shown, the input module may include but is not limited to:

[0056] A first input unit connected to an input terminal of the learning vector quantization module to input a stable parameter into the learning vector quantization module;

[0057] A second input unit is connected to an input terminal of the Bayesian module to input the statistical index into the Bayesian module;

[0058] The third input unit is connected to the input end of the deep learning neural network module to input the fluctuation parameter, the stability parameter and the statistical indicators of each fluctuation parameter into the deep learning neural network module.

[0059] In this embodiment, the characteristics of the three types of input data - fluctuation parameters, stable parameters, and statistical indicators are further taken into consideration, and the learning vector quantization module is selected to process only the stable parameters to map the points in the data set to a discrete point set; the Bayesian module is selected to process only the statistical parameters as a statistical method; and the deep learning neural network is selected to process the fluctuation parameters, stable parameters, and statistical indicators of each fluctuation parameter to seek the intricate correlations within each data and among the three types of data, which can give full play to the strengths of each module, improve the overall prediction accuracy of the model, reduce data redundancy and processing volume, and reduce the time and pressure of model calculation.

[0060] (2) Vector quantization module, Bayesian module and deep learning neural network module:

[0061] More specifically, the specific structural forms of the learning vector quantization module, the Bayesian module and the deep learning neural network module can be set arbitrarily.

[0062] Preferably, Figure 2 , 4 As shown, the learning vector quantization module can be selected as a hidden layer. The number of hidden nodes, the initial position of the hidden nodes, and the weights of each competitive layer can be set arbitrarily and iteratively trained. As shown in the above example, the input stable parameters are: equipment service life, equipment historical fault records, equipment maintenance records, equipment production years, equipment model, equipment configuration, a total of 6, using a6-a 11 Represents, and after passing through the hidden layer, outputs a one-dimensional feature vector (q1,.....q7);

[0063] The Bayesian module can be represented by the function f(x). The definition of the f(x) function is: Let the statistical indicators: the difference, amplitude, and phase of temperature, pressure, vibration, current, and voltage within a period of time, a total of 15, and use a 12 -a 26 Meaning: x={a 12 ,....a 26} is an item to be classified, and each a is a characteristic attribute of x; there is a category set C = {z,g}, z1 represents normal, g2 represents fault; calculate f(x)=(t1,t2), t1=P(z|x), is the probability of normal equipment, t2=P(g|x) is the probability of equipment fault.

[0064] A deep neural network module may optionally include three layers of convolutional units connected in sequence and a pooling layer connected at the end; each convolutional unit includes a convolution kernel and a binary classification activation function, such as a sigmoid binary classification function; for example, Figure 5As shown, the first convolution unit has a convolution kernel of 3×3, a step size of 3, and a padding of 2; the second convolution unit has a convolution kernel of 5×5, a step size of 4, and a padding of 2; the third convolution unit has a convolution kernel of 7×7, a step size of 5, and a padding of 2, and outputs a one-dimensional feature vector (p1, ...p7). More preferably, as Figure 2 As shown, before the convolution layer, a normalization layer may be optionally included to pull the data of the feature matrix to a standard normal distribution with a mean of 0 and a variance of 1. In this embodiment, a labeling layer is set to adjust the data distribution to simplify the difficulty of data calculation and improve accuracy. Specifically, the normalization formula is:

[0065] x'= =(x-mean(x)) / std(x)

[0066] Among them, x' is the processed data, mean(x) is the mean of the data, and std(x) is the variance of the data. After processing, the data obeys the standard normal distribution.

[0067] In this embodiment, the preferred structures of the learning vector quantization module, the Bayesian module and the deep learning neural network module are provided to adapt to the input and output of each module. It is worth noting that the specific structure of each module is not limited to this, but is only a preferred example.

[0068] (3) Output module:

[0069] More preferably, in one embodiment, Figure 2 As shown, in the output module, the learning vector quantization module, the Bayesian module and the deep learning neural network module can be configured with their own fully connected layers, and then the superposition module is linked, and the total probability obtained by superposition is the final output result of the prediction model.

[0070] Specifically, the learning vector quantization module outputs the first probability y1 (q1, q2); q1 is the first normal probability, q2 is the first fault probability; the deep learning neural network module outputs the second probability y2 (r1, r2), r1 is the second normal probability, r2 is the second fault probability; the Bayesian module outputs the third probability y3 (t1, t2), t1 is the third normal probability, t2 is the third fault probability. Finally, the superposition module outputs the total normal probability R and the total fault probability R':

[0071] R = q1+r1+t1

[0072] R'=q2+r2+t2

[0073] More preferably, in another embodiment, Figure 4As shown, in the output module, you can choose to first superimpose the outputs of the learning vector quantization module and the deep learning neural network module, then link the fully connected layer to obtain the total output of the vector quantization module and the deep learning neural network module, and then link the superposition module to superimpose the output of the Bayesian module. The total probability obtained by superposition is the final output result of the prediction model.

[0074] Specifically, the learning vector quantization module outputs a one-dimensional feature vector (q1, .....q7) after the hidden layer, and the deep learning neural network module outputs a one-dimensional feature vector (p1, ...p7) after three convolutional layers. After the fully connected layer, the first superposition probability y4 is obtained, which is then superimposed with the third probability y3 output by the Bayesian module to obtain the second superposition probability as the total normal probability and the total fault probability.

[0075] In this embodiment, many preferred embodiments of the prediction model are given. In the input module, the respective advantages of the three modules are preferably brought into play, and specific data input corresponding modules are selected respectively. Under the premise of not affecting the prediction accuracy, data redundancy and model prediction time can be further reduced; in the vector quantization module, the Bayesian module and the deep learning neural network module, the specific structural form of each module is given; in the output module, two preferred embodiments are given. In the first embodiment, the outputs of the three modules are superimposed to obtain the total output, and the type with a large probability - normal or fault is comprehensively selected as the prediction result, which can combine the advantages of the three models, give play to the comprehensiveness of the data of the fluctuation parameters, stability parameters and statistical indicators, and improve the prediction accuracy; in the second embodiment, based on the principle of preferential combination of homogeneous data, the correlation between the outputs of the vector quantization module and the deep learning neural network module is first sought through the fully connected layer to obtain the first superimposed probability, and then combined with the output of the Bayesian module, the output result is more targeted and accurate.

[0076] B: Prediction model training steps:

[0077] The running data of the training data set constructed in step S1 is input into the prediction model to obtain the prediction result, corresponding to the annotation result, and the parameters of the prediction model are iteratively updated until the prediction accuracy reaches a preset threshold, and the iteration is stopped to obtain the trained prediction model. Preferably, model evaluation and optimization are also included: the model is evaluated by methods such as cross-validation, and the model is optimized and adjusted according to the evaluation results to improve the prediction accuracy and reliability of the model.

[0078] S3: Collect the current operation data of the equipment, including fluctuation parameters and stability parameters; and calculate the statistical indicators of each fluctuation parameter;

[0079] Specifically, the current operation data of the device may be collected in the same manner as step S1. Similarly, it is preferably represented by a current feature matrix.

[0080] S4: Input the current fluctuation parameters, stability parameters and statistical indicators of each fluctuation parameter into the trained prediction model, output the normal and failure probabilities of the equipment, and obtain the failure prediction results.

[0081] Specifically, the current feature matrix is ​​input into the trained prediction model to obtain the normal probability and fault probability of the equipment, and the normal threshold and / or fault threshold can be optionally set; when the normal probability is greater than the normal threshold, the equipment is predicted to be in a normal state, otherwise the equipment is faulty; or when the fault probability is greater than the fault threshold, the equipment is predicted to be in a faulty state, otherwise the equipment is normal.

[0082] Specifically, the current feature matrix X is passed through the learning vector quantization module to output the first probability y1 (q1, q2); q1 is the first normal probability, q2 is the first fault probability; the deep learning neural network module outputs the second probability y2 (r1, r2), r1 is the second normal probability, r2 is the second fault probability; the Bayesian module outputs the third probability y3 (t1, t2), t1 is the third normal probability, t2 is the third fault probability. Finally, through the superposition module, the total normal probability R and the total fault probability R' are output:

[0083] R = q1+r1+t1

[0084] R'=q2+r2+t2

[0085] Example:

[0086] When R' is greater than or equal to 2, it means that the device prediction result is a fault; otherwise, it means that the device prediction result is not a fault, that is, normal.

[0087] Alternatively, the current feature matrix X outputs a one-dimensional feature vector (q1, .....q7) through the learning vector quantization module, and outputs a one-dimensional feature vector (p1, ...p7) through the deep learning neural network module. After the full connection layer, the first superposition probability y4 is obtained, and then superimposed with the third probability y3 output by the Bayesian module to obtain the second superposition probability as the total normal probability and the total fault probability. If the total normal probability is greater than 0.65, it means that the device prediction result is normal, otherwise it means that the device prediction result is faulty.

[0088] More preferably, the optimized model can be used to monitor and predict the operating status of the equipment in real time, which also includes: when the prediction results show that the equipment may fail, an early warning message is issued to promptly notify relevant personnel to perform maintenance; or according to the fault prediction results and early warning information, a corresponding maintenance strategy is formulated, such as planned maintenance, preventive maintenance, fault repair, etc., to reduce the risk of equipment failure and improve the reliability and availability of the equipment.

[0089] In addition, the following options are available: Continuous improvement: Regularly evaluate and optimize the predictive maintenance system, and adjust and improve data collection, feature extraction, model building, and maintenance strategies according to actual conditions to improve the effectiveness and efficiency of predictive maintenance. Because the specific implementation of predictive maintenance methods may vary depending on the type of equipment, data characteristics, and maintenance requirements, they need to be adjusted and optimized according to specific circumstances. At the same time, predictive maintenance also needs to fully consider the actual operation of the equipment and maintenance costs to ensure the feasibility and economy of the maintenance strategy.

[0090] This patented predictive maintenance method can achieve the following benefits:

[0091] Reduce equipment failure rate: Through real-time monitoring and analysis of the equipment's operating status, based on historical and current operating data, predict possible equipment failures, take maintenance measures in advance, and reduce equipment failure rate;

[0092] Reduce maintenance costs: Predictive maintenance can avoid emergency repairs after equipment failure occurs, reducing maintenance costs and production losses;

[0093] Improve production efficiency: Predictive maintenance can avoid production line downtime caused by equipment failure and improve production efficiency;

[0094] Optimize equipment maintenance strategy: Predictive maintenance can optimize equipment maintenance strategy according to the actual operation of the equipment, and achieve more refined and intelligent equipment maintenance.

[0095] Improve equipment reliability and availability: Predictive maintenance can detect equipment failures and anomalies in a timely manner, improve equipment reliability and availability, and ensure the stability and continuity of production operations.

[0096] In summary, the present invention provides a fault prediction method, and the fault predictive maintenance is a maintenance strategy that uses advanced data analysis technology and machine learning algorithms to predict possible future failures or performance degradation of equipment by monitoring the operating status, historical data and other relevant factors of the equipment, so as to take preventive maintenance measures before the failure occurs; specifically, in terms of foresight and feasibility, a feature matrix is ​​constructed as the input of a prediction model in combination with its fluctuation parameters, stability parameters and statistical indicators, which is comprehensive, and the prediction model based on deep neural network, vector quantization and Bayesian three-form structure is adopted, which can further comprehensively extract the unique characteristics of fluctuation parameters, stability parameters and statistical indicators, and then comprehensively consider and superimpose the outputs of the three modules as the total output of the prediction model, thereby improving the prediction accuracy of the prediction model.

[0097] On the other hand, the present invention further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned fault prediction methods.

[0098] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores program codes that can be executed by the processor; the program codes are used to execute any of the above-mentioned fault prediction methods.

[0099] Exemplarily, the program code may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the program code in the terminal device.

[0100] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the terminal device may also include an input / output device, a network access device, a bus, etc.

[0101] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0102] The memory may be an internal storage unit of the terminal device, such as a hard disk or a memory. The memory may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal device. Further, the memory may also include both an internal storage unit of the terminal device and an external storage device. The memory is used to store the program code and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is to be output.

[0103] The above-mentioned computer storage medium and terminal device are created based on the above-mentioned fault prediction method, and their technical functions and beneficial effects are no longer repeated here. The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the various technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A fault prediction method, It is characterized in that include: Collect historical operation data of the equipment under normal and fault conditions, including fluctuation parameters, stable parameters and statistical indicators of each fluctuation parameter, and mark them as normal data or fault data to build a training data set; Construct a prediction model with fluctuation parameters, stability parameters and statistical indicators of each fluctuation parameter as input and equipment normal and failure probabilities as output; and obtain a trained prediction model based on the training data set; Collect the current operating data of the equipment, including fluctuation parameters and stability parameters; and calculate the statistical indicators of each fluctuation parameter; The current fluctuation parameters, stability parameters and statistical indicators of each fluctuation parameter are input into the trained prediction model, and the normal and failure probabilities of the equipment are output to obtain the failure prediction results; Prediction model, including: input module, learning vector quantization module, Bayesian module, deep learning neural network module and output module; The input ends of the learning vector quantization module, the Bayesian module and the deep learning neural network module are connected to the input module, and the output ends are connected to the input end of the output module; the output probabilities of the learning vector quantization module, the Bayesian module and the deep learning neural network module are combined to output the probabilities of normal and faulty devices; Input modules, including: A first input unit connected to an input terminal of the learning vector quantization module to input a stable parameter into the learning vector quantization module; A second input unit is connected to an input terminal of the Bayesian module to input the statistical index into the Bayesian module; The third input unit is connected to the input end of the deep learning neural network module to input the fluctuation parameter, the stability parameter and the statistical indicators of each fluctuation parameter into the deep learning neural network module.

2. The fault prediction method according to claim 1, It is characterized in that In the output module, the learning vector quantization module, the Bayesian module, and the deep learning neural network module are configured with their own fully connected layers, and then the superposition module is linked to obtain the total probability as the final output result of the prediction model; Or, in the output module, first superimpose the outputs of the learning vector quantization module and the deep learning neural network module, then link the fully connected layer to obtain the total output of the vector quantization module and the deep learning neural network module, then link the superposition module, superimpose the output of the Bayesian module, and superimpose the total probability obtained as the final output result of the prediction model.

3. The fault prediction method according to claim 1, It is characterized in that The deep learning neural network module includes three layers of convolutional units connected in sequence and a pooling layer at the end; each convolutional unit includes a layer of convolution kernel and a binary classification activation function.

4. The fault prediction method according to claim 3, It is characterized in that The first convolution unit has a convolution kernel of 3×3, a step size of 3, and a padding of 2; the second convolution unit has a convolution kernel of 5×5, a step size of 4, and a padding of 2; the third convolution unit has a convolution kernel of 7×7, a step size of 5, and a padding of 2; finally, after the fully connected layer, the first probability y1(q1,q2) is output; q1 is the first normal probability, and q2 is the first fault probability.

5. The fault prediction method according to claim 1, It is characterized in that When the failure probability is greater than or equal to the failure threshold, it means that the device prediction result is a failure; otherwise, it means that the device prediction result is normal.

6. The fault prediction method according to any one of claims 1 to 5, It is characterized in that Fluctuating parameters, including any one or more of temperature, pressure, vibration, current, and voltage; Stability parameters, including any one or more of equipment service life, equipment historical failure records, equipment maintenance records, equipment production years, equipment model, and equipment configuration; Statistical indicators include any one or more of the difference, amplitude, and phase of each fluctuation parameter.

7. A computer storage medium, It is characterized in that An executable program code is stored; the executable program code is used to execute the fault prediction method according to any one of claims 1 to 6.

8. A terminal device, It is characterized in that It comprises a memory and a processor; the memory stores program codes executable by the processor; the program codes are used to execute the fault prediction method according to any one of claims 1 to 6.

Images