Equipment fault prediction method and device
By designing equipment fault prediction devices, using time series analysis and ARIMA models to predict future faults, the problem of limited prediction time in the prior art is solved, and the practicality and accuracy of fault prediction are improved.
Patent Information
- Application Number
- CN202510281409.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-11
- Publication Date
- 2025-06-27
AI Technical Summary
The prior art can only obtain current signals or data in equipment failure prediction, and cannot predict the situation in the next few seconds or period of time, resulting in limited prediction time, poor practicality, and inability to maintain it in time.
A device fault prediction device is designed, including a normal training module, a device signal collection module, a current fault prediction module, a time series analysis module and a future fault prediction module. By collecting historical normal data of the device and current operating data, performing feature extraction and database construction, continuously recording and analyzing device operating data, and using the ARIMA model to predict future failures.
It realizes the prediction of equipment failures in the next few seconds or period of time, improves the time range and practicality of the prediction, ensures that staff can carry out timely maintenance and ensures the normal operation of the equipment.
Smart Images
Figure CN120217231A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault prediction, and particularly to a method and device for predicting equipment faults. Background Art
[0002] Currently, during the use of equipment, as time goes by and the number of uses accumulates, it is inevitable that equipment faults will occur. Once a fault occurs, it will affect the normal operation of the production line or device, and further affect the work progress. Therefore, in actual use, in order to ensure the long-term stable operation of the equipment, it is necessary to predict equipment faults. After predicting a fault, maintenance or replacement should be carried out in a timely manner to effectively reduce losses.
[0003] However, in the aforementioned prior art, when currently predicting equipment faults, usually only the current signals or data of the equipment are obtained to predict the current faults, and the situation in the next few seconds or a period of time cannot be predicted. Therefore, the prediction time is limited and the practicality of the prediction is poor, resulting in the inability of the staff to reach the equipment in time for maintenance before the fault occurs. Summary of the Invention
[0004] The purpose of the present invention is to provide a method and device for predicting equipment faults, which solve the problem that in the prior art, when currently predicting equipment faults, usually only the current signals or data of the equipment are obtained to predict the current faults, and the situation in the next few seconds or a period of time cannot be predicted. Therefore, the prediction time is limited and the practicality of the prediction is poor, resulting in the inability of the staff to reach the equipment in time for maintenance before the fault occurs.
[0005] To achieve the above purpose, the present invention provides a device for predicting equipment faults, including a normal training module, an equipment signal collection module, a current fault prediction module, a time series analysis module, and a future fault prediction module; the normal training module, the equipment signal collection module, the current fault prediction module, the time series analysis module, and the future fault prediction module are connected in sequence;
[0006] The normal training module is used to collect the historical normal data of the equipment to obtain a normal database of the equipment;
[0007] The equipment signal collection module is used to collect the current operation data of the equipment;
[0008] The current fault prediction module is used to compare the newly collected operation data with the normal data in the normal database. When a difference value appears, it indicates that the current equipment has a fault;
[0009] The time series analysis module is used to continuously record the operation data trajectory of the equipment and perform time series analysis to obtain a time series analysis result;
[0010] The future fault prediction module is used to perform future fault prediction based on the time series analysis result to obtain a future fault prediction result.
[0011] Among them, the normal training module includes a feature extraction unit and a database construction unit, and the feature extraction unit is connected to the database construction unit;
[0012] The feature extraction unit is used to query logs and servers, collect the historical normal data, and extract features therefrom to obtain high-value features;
[0013] The database construction unit is used to store the high-value features in a database, and mark the corresponding faults that occur when the features are abnormal on each of the high-value features, and finally obtain a device normal database.
[0014] Among them, the feature extraction unit includes a spectrum analysis subunit, a wavelet transform subunit, and a principal component analysis subunit, and the spectrum analysis subunit, the wavelet transform subunit, and the principal component analysis subunit are connected in sequence;
[0015] The spectrum analysis subunit is used to convert the signal from the time domain to the frequency domain, calculate the spectrum using Fourier transform or fast Fourier transform, and obtain the audio and vibration features of the device;
[0016] The wavelet transform subunit is used to perform multi-scale decomposition on the signal, extract features using discrete wavelet transform or continuous wavelet transform, and obtain the frequency band features of the device signal;
[0017] The principal component analysis subunit is used to perform dimensionality reduction processing on high-dimensional data, extract the main feature components or discriminative features, and package all the extracted features to obtain high-value features.
[0018] Among them, the device signal collection module includes a data collection unit and a data preprocessing unit, and the data collection unit is connected to the data preprocessing unit;
[0019] The data collection unit is used to collect the current operating data through sensors installed on the device, and the operating data includes temperature, vibration frequency, and pressure;
[0020] The data preprocessing unit is used to clean and transform the collected operating data.
[0021] Among them, the time series analysis module includes a device data continuous recording unit, a trend analysis unit, a seasonal analysis unit, and an autocorrelation analysis unit, and the device data continuous recording unit, the trend analysis unit, the seasonal analysis unit, and the autocorrelation analysis unit are connected in sequence;
[0022] The device data continuous recording unit is used to continuously collect the operation data of the device, obtain the device continuous data, and visually display the device continuous data using a time series graph;
[0023] The trend analysis unit is used to smooth the device continuous data using a moving average and identify the long-term trend in the data;
[0024] The seasonal analysis unit is used to extract and analyze the seasonal components of the device continuous data using seasonal decomposition to obtain seasonal analysis results;
[0025] The autocorrelation analysis unit is used to calculate the autocorrelation coefficient of the device continuous data, understand the correlation data at different time lags, and package the correlation data, the long-term trend, and the seasonal analysis results to obtain time series analysis results.
[0026] Among them, the future fault prediction module includes a data stationarity test unit, an autocorrelation unit, an ARIMA model optimization unit, and a future prediction unit, and the data stationarity test unit, the autocorrelation unit, the ARIMA model optimization unit, and the future prediction unit are connected in sequence;
[0027] The data stationarity test unit processes the time series analysis results through an ARIMA model and uses the ADF test method to determine whether the sequence is stationary;
[0028] The autocorrelation unit is used to plot the autocorrelation function and partial autocorrelation function graphs of the time series data, and preliminarily determine the autoregressive order p and moving average order q of the ARIMA model by analyzing the ACF and PACF graphs;
[0029] The ARIMA model optimization unit is used to optimize, fit, and diagnose the ARIMA model;
[0030] The future prediction unit is used to predict future values using the fitted ARIMA model, set different prediction steps to obtain predicted values at different future time points, and obtain future fault prediction results.
[0031] Among them, the ARIMA model optimization unit includes an optimal parameter selection subunit, a model fitting subunit, and a model diagnosis subunit, and the optimal parameter selection subunit, the model fitting subunit, and the model diagnosis subunit are connected in sequence;
[0032] The optimal parameter selection subunit is used to automatically find the optimal combination of p, d, q parameters using information criteria, and then weigh the complexity and fitting degree of the model to select the best model parameters;
[0033] The fitting model subunit is used to fit an ARIMA model using the determined parameters p, d, and q;
[0034] The model diagnosis subunit is used to determine whether the residual sequence is a white noise sequence through the ACF and PACF plots of the residual sequence and the Ljung-Box Q test.
[0035] The present invention also provides a device fault prediction method, which employs the above-mentioned device fault prediction device and includes the following steps:
[0036] By collecting the historical normal data of the device, a normal database of the device is obtained;
[0037] Compare the newly collected operation data with the normal data in the normal database. When there is a difference value, it indicates that the current device has a fault;
[0038] Continuously collect the operation data of the device to obtain continuous device data, and visually display the continuous device data using a time series graph;
[0039] Through trend, seasonal, and autocorrelation analysis, a time series analysis result is obtained;
[0040] Input the time series analysis result into the ARIMA model for processing, and optimize the ARIMA model at the same time;
[0041] Set different prediction steps through the ARIMA model to obtain future fault prediction results.
[0042] For a device fault prediction method and device of the present invention, the normal training module is used to collect the historical normal data of the device to obtain a normal database of the device; the device signal collection module is used to collect the current operation data of the device; the current fault prediction module is used to compare the newly collected operation data with the normal data in the normal database. When there is a difference value, it indicates that the current device has a fault; the time series analysis module is used to continuously record the operation data trajectory of the device and perform time series analysis to obtain a time series analysis result; the future fault prediction module is used to perform future fault prediction based on the time series analysis result to obtain future fault prediction results;
[0043] Thus, first predict the current device faults. If it is predicted that there are current faults or upcoming faults, then perform maintenance. If it is predicted that the device has no faults, collect historical data to predict the future faults of the device, so that the staff can understand whether there are faults in the device in the future for a period of time, which is beneficial for the staff to perform maintenance in a timely manner, greatly ensure the normal operation of the device, avoid device operation interruption, and greatly improve the practicality. Description of the Drawings
[0044] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art.
[0045] Figure 1 It is the schematic diagram of the device fault prediction device of the present invention.
[0046] Figure 2 It is the schematic diagram of the normal training module of the present invention.
[0047] Figure 3 It is the schematic diagram of the feature extraction unit of the present invention.
[0048] Figure 4 It is the schematic diagram of the device signal collection module of the present invention.
[0049] Figure 5 It is the schematic diagram of the time series analysis module of the present invention.
[0050] Figure 6 It is the schematic diagram of the future fault prediction module of the present invention.
[0051] Figure 7 It is the schematic diagram of the ARIMA model optimization unit of the present invention.
[0052] Figure 8 It is the step flowchart of the device fault prediction method of the present invention.
[0053] 1 - Normal training module, 101 - Feature extraction unit, 1011 - Spectrum analysis subunit, 1012 - Wavelet transform subunit, 1013 - Component analysis subunit, 102 - Database construction unit, 2 - Device signal collection module, 201 - Data collection unit, 202 - Data preprocessing unit, 3 - Current fault prediction module, 4 - Time series analysis module, 401 - Device data continuous recording unit, 402 - Trend analysis unit, 403 - Seasonal analysis unit, 404 - Autocorrelation analysis unit, 5 - Future fault prediction module, 501 - Data stationarity test unit, 502 - Autocorrelation unit, 503 - ARIMA model optimization unit, 5031 - Select optimal parameter subunit, 5032 - Fit model subunit, 5033 - Model diagnosis subunit, 504 - Future prediction unit. Detailed Description of the Embodiments
[0054] The following will describe in detail the embodiments of the present invention. The examples of the embodiments are shown in the drawings. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.
[0055] Please refer to Figures 1 to 7 , the present invention provides a device fault prediction device, specifically including:
[0056] The normal training module 1 is used to collect the historical normal data of the device to obtain the normal database of the device;
[0057] Specifically including:
[0058] The feature extraction unit 101 is used to query the log and the server, collect the historical normal data, and extract features therefrom to obtain high-value features;
[0059] Specifically including:
[0060] The spectrum analysis subunit 1011 is used to convert the signal from the time domain to the frequency domain, calculate the spectrum using Fourier transform or fast Fourier transform, and obtain the audio and vibration characteristics of the device;
[0061] First, the staff collects the historical normal data of the device by querying the device log and server records, and then performs spectrum analysis on the data signal. By converting the audio signal from the time domain to the frequency domain, the energy distribution of the audio signal at different frequencies can be clearly seen, thereby extracting the spectrum characteristics of the audio; for the vibration signal, frequency domain analysis is equally important. Through algorithms such as FFT, the vibration signal can be decomposed into sine wave components of different frequencies, thereby identifying the dominant frequency and harmonic components in the vibration signal; this helps to diagnose the operating state of mechanical equipment, such as bearing faults, imbalance, etc., and further achieve predictive maintenance.
[0062] The wavelet transform subunit 1012 is used to perform multi-scale decomposition on the signal, extract features using discrete wavelet transform or continuous wavelet transform, and obtain the frequency band characteristics of the device signal;
[0063] Multi-scale decomposition decomposes the signal into sub-signals at different scales. At each scale, the signal is decomposed into a low-frequency part (approximate signal) and a high-frequency part (detail signal). The low-frequency part reflects the main trend of the signal, while the high-frequency part contains the detailed information of the signal; the discrete wavelet transform (DWT) is a discretized form of the continuous wavelet transform. It extracts the local features of the signal at different scales through multi-scale and multi-position analysis of the signal; the continuous wavelet transform (CWT) represents the signal as a continuous integral of wavelet coefficients, which can provide the localized information of the signal in the continuous time and frequency domains.
[0064] The principal component analysis subunit 1013 is used to perform dimensionality reduction processing on the high-dimensional data, extract the main feature components or discriminative features, and package all the extracted features to obtain high-value features.
[0065] High-dimensional data usually contains a large number of features, which will increase the computational complexity in subsequent analysis and modeling processes. By relying on dimensionality reduction processing, the number of features can be reduced, thereby reducing the computational overhead and improving the processing speed. Therefore, the sampling principal component analysis (PCA) method is adopted to map high-dimensional data to a low-dimensional space through linear transformation. After extracting the main feature components or discriminative features, these features can be packed. The purpose of packing is to combine multiple features into a high-value feature set for subsequent analysis and modeling tasks. The packing process may involve operations such as feature standardization, normalization, and weighting to improve the quality and usability of the feature set.
[0066] The database construction unit 102 is used to store the high-value features in the database and mark the corresponding faults that occur when each of the high-value features is abnormal, and finally obtain a device normal database.
[0067] Thus, by storing the high-value features in the database and marking the corresponding faults that occur when each feature is abnormal, once a new abnormal feature is detected subsequently, the system can directly compare it with the information recorded in the database to immediately discover the corresponding fault.
[0068] The device signal collection module 2 is used to collect the current operating data of the device;
[0069] Specifically including:
[0070] The data collection unit 201 is used to collect the current operating data through sensors installed on the device, and the operating data includes temperature, vibration frequency, and pressure;
[0071] After relying on sensors to collect data, it is convenient for subsequent data preprocessing.
[0072] The data preprocessing unit 202 is used to clean and transform the collected operating data.
[0073] During data cleaning, identify and delete invalid, duplicate, missing, or incorrect data records to ensure data quality and accuracy. For missing data, the cleaning process may involve filling in missing values (such as using the mean, median, mode, or specific algorithms for predictive filling), or deleting records containing missing values based on business logic. During data transformation: it involves standardizing the data (such as scaling features to a specific range) or normalizing the data (such as transforming the data into a distribution with a mean of 0 and a variance of 1) to eliminate the dimensional differences between different features and improve the performance of the model; it includes feature selection (selecting the most useful features for the model), feature construction (combining multiple features to generate new features), and feature extraction (such as using methods like PCA, LDA, etc. to extract more representative features from the original features); through data cleaning and transformation, the quality and usability of the data can be significantly improved, thereby optimizing the performance of subsequent data analysis and machine learning tasks. These steps help reduce noise, eliminate biases, improve the representativeness and generalization ability of the data, and ultimately enhance the accuracy and reliability of the model.
[0074] The current fault prediction module 3 is used to compare the newly collected operation data with the normal data in the normal database. When there is a difference value, it indicates that the current device has a fault.
[0075] The time series analysis module 4 is used to continuously record the operation data trajectory of the device and perform time series analysis to obtain the time series analysis result.
[0076] Specifically, it includes:
[0077] The device data continuous recording unit 401 is used to continuously collect the operation data of the device to obtain the device continuous data, and visually display the device continuous data using a time series graph.
[0078] By continuously collecting data and visually displaying it, it helps the staff to observe the data in real time, facilitating the understanding of the latest device operation data at any time. At the same time, the device continuous data is also used for subsequent predictive analysis.
[0079] The trend analysis unit 402 is used to smooth the device continuous data using a moving average to identify the long-term trend in the data.
[0080] The moving average can effectively smooth the short-term fluctuations of the data, making the long-term trend of the data more obvious. This is very helpful for identifying the long-term change trend of device data; by observing the trend of the moving average, the long-term trend of device data can be identified. This is of great significance for predicting the future operation state of the device, formulating maintenance plans, and optimizing production plans, etc. And the data processed by the moving average is smoother and more stable, which helps to improve the data quality and the accuracy of the analysis results.
[0081] The seasonal analysis unit 403 is used to extract and analyze the seasonal component of the device continuous data using seasonal decomposition to obtain a seasonal analysis result;
[0082] Seasonal decomposition can adopt an additive model or a multiplicative model. In the additive model, the time series is represented as the sum of a trend, seasonality, and random residuals; while in the multiplicative model, the time series is represented as the product of these components. Which model to choose depends on the characteristics of the data. If the amplitude of the seasonal fluctuations in the data is relatively stable, the additive model can be used; if the amplitude of the seasonal fluctuations changes over time, the multiplicative model can be used; in actual operation, seasonal decomposition is usually achieved with the help of statistical software or data analysis tools; automatically calculate and extract the seasonal component in the time series and visually display it in the form of a chart; finally, by observing these charts, the staff can deeply analyze the seasonal fluctuation law of the data and provide a basis for subsequent decision-making and management;
[0083] Therefore, seasonal decomposition can extract the seasonal component from the device data, thereby visually identifying the periodic fluctuations of the data at different time points (such as quarters, months). This is crucial for understanding the seasonal operation law of the device, predicting future states, and formulating targeted maintenance and management strategies.
[0084] The autocorrelation analysis unit 404 is used to understand the correlation data of the data at different time lags by calculating the autocorrelation coefficient of the device continuous data, and package the correlation data, the long-term trend, and the seasonal analysis result to obtain a time series analysis result.
[0085] Package all the analysis results for subsequent future fault prediction.
[0086] The future fault prediction module 5 is used to perform future fault prediction based on the time series analysis result to obtain a future fault prediction result.
[0087] Specifically include:
[0088] The data stationarity test unit 501 processes the time series analysis result through an ARIMA model and uses the ADF test method to determine whether the sequence is stationary;
[0089] If the sequence is not stationary, it is necessary to make it stationary through methods such as differencing or logarithmic differencing; determine the differencing order; according to the stationarity test result, determine the differencing order d. If the sequence is not stationary, differencing processing is required until the sequence becomes stationary.
[0090] The autocorrelation unit 502 is used to plot the autocorrelation function and partial autocorrelation function graphs of time series data. By analyzing the ACF and PACF graphs, the autoregressive order p and moving average order q of the ARIMA model are preliminarily determined.
[0091] The ACF graph shows the correlation between time series data and its lagged values, which helps to identify periodicity or trends in the data. The PACF graph excludes the influence of other lagged values and focuses on measuring the direct relationship between a certain moment and its lagged value, thereby identifying data characteristics. At the same time, by analyzing the shape and trend of the ACF and PACF graphs, the p (autoregressive order) and q (moving average order) parameters of the ARIMA model can be preliminarily determined. The accurate selection of p and q parameters is crucial for constructing an effective prediction model.
[0092] The ARIMA model optimization unit 503 is used to optimize, fit, and diagnose the ARIMA model.
[0093] Specifically, it includes:
[0094] The optimal parameter selection subunit 5031 is used to automatically find the optimal combination of p, d, and q parameters using information criteria, and then balance the complexity and goodness of fit of the model, so as to select the best model parameters.
[0095] First, define the value ranges of the p, d, and q parameters, that is, the parameter space. This space can be a discrete grid or a continuous interval. Then, by traversing or sampling different points in the parameter space (i.e., different combinations of p, d, and q parameters), construct the corresponding ARIMA models. Secondly, for each constructed model, calculate its AIC or BIC value. These values reflect the goodness of fit and complexity of the model.
[0096] By adopting the Akaike information criterion (AIC), the goodness of fit of the model to the data is measured by calculating the sum of squared residuals (or other loss functions) of the model. A smaller AIC value means a better fit of the model to the data; AIC also takes into account the number of parameters in the model and imposes a penalty on models with a larger number of parameters to avoid overfitting. This helps to find a balance between the goodness of fit and complexity of the model.
[0097] AIC is based on the concept of information entropy and selects the model by maximizing the posterior probability of the model (or equivalently, minimizing the information loss); it takes into account the goodness of fit of the model (measured by the likelihood function) and the number of parameters (as a measure of complexity). The formula for AIC is usually: AIC = 2k - 2ln(L), where k is the number of parameters in the model and L is the maximum likelihood estimate of the model.
[0098] The fitting model subunit 5032 is used to fit an ARIMA model using the determined parameters p, d, and q;
[0099] The fitting process constructs an ARIMA model with the given parameters p, d, and q; p represents the number of autoregressive terms, d represents the number of differencing times, and q represents the number of moving average terms. Through fitting, the specific values of these parameters can be determined, thus constructing a model that matches the equipment failure data for subsequent future equipment failure prediction.
[0100] The model diagnosis subunit 5033 is used to determine whether the residual sequence is a white noise sequence through the ACF and PACF plots of the residual sequence and the Ljung-Box Q test.
[0101] The residual sequence is the difference between the model predicted value and the actual observed value. If it behaves as white noise, that is, there is no significant correlation between the values in the sequence, it indicates that the ARIMA model has captured the main trends and patterns in the data, and the model fitting effect is good; a white noise sequence means that there are no exploitable dynamic laws in the sequence, that is, the model has fully extracted the useful information in the data without missing important patterns or trends; at the same time, if the residual sequence is white noise, then the prediction based on the ARIMA model will be more reliable because the model has accurately described the data generation process, and future predicted values will be more likely to be close to the actual observed values;
[0102] The ACF plot shows the correlation between the residual sequence and its lagged versions. If all autocorrelation coefficients in the ACF plot are close to zero and fluctuate within the confidence interval, then the residual sequence can be considered white noise; the PACF plot shows the correlation between the residual sequence and its own lagged version after considering the influence of other lagged versions. Similarly, if all partial autocorrelation coefficients in the PACF plot are close to zero and fluctuate within the confidence interval, then the residual sequence can also be considered white noise;
[0103] In the Ljung-Box Q test, by calculating the test statistic and comparing it with the critical value, it can be determined whether there is significant autocorrelation in the residual sequence. If the statistic is not significant (i.e., the p-value is greater than the significance level), the null hypothesis that the residual sequence is white noise cannot be rejected. The Ljung-Box Q test can examine the autocorrelation of the residual sequence at multiple lag orders, thus more comprehensively evaluating the model fitting effect. If there is no significant autocorrelation in the residual sequence at all considered lag orders, then it can be more certain that the residual sequence is white noise. Thus, through a variety of means, it is ultimately determined whether the residual sequence is a white noise sequence, ensuring the reliability of the ARIMA model. If it is detected that it is not white noise, it means that there may still be information or structure in the model that has not been fully captured, or the model may be improperly set. Then, it is necessary to re-evaluate and adjust the parameters (p, d, q) of the ARIMA model, try different combinations to improve the model fitting, or re-conduct the stationarity test.
[0104] The future prediction unit 504 is used to predict future values using the fitted ARIMA model, set different prediction steps to obtain predicted values at different future time points, and obtain the future fault prediction result.
[0105] During the prediction process, the ARIMA model fits a model that can describe the data generation process based on historical data, and then uses this model to predict future data; the setting of the prediction step determines how far into the future the model needs to predict data; if the prediction step is set to 1, the model will predict the data at the next time point; if the prediction step is set to n, the model will predict the data at the nth future time point; thus, understanding the future trend and changes of the time series data, and finally obtaining the future fault prediction result for the staff to view; enabling the staff to understand whether there is a fault in the device in the future for a period of time, which is conducive to the staff to carry out maintenance in a timely manner, greatly ensuring the normal operation of the device and avoiding the interruption of device operation.
[0106] Please refer to Figure 8 The present invention also provides a device fault prediction method, including the following steps:
[0107] S1: By collecting the historical normal data of the device, a normal database of the device is obtained;
[0108] S2: Compare the newly collected operation data with the normal data in the normal database. When there is a difference value, it indicates that the current device has a fault;
[0109] S3: Continuously collect the operation data of the device to obtain continuous device data, and visually display the continuous device data using a time series graph;
[0110] S4: Obtain the time series analysis result through trend, seasonality, and autocorrelation analysis;
[0111] S5: Input the time series analysis result into the ARIMA model for processing, and optimize the ARIMA model simultaneously;
[0112] S6: Set different prediction steps through the ARIMA model to obtain the future fault prediction result.
[0113] Among them, by collecting the historical normal data of the device, the normal database of the device is obtained; compare the newly collected operation data with the normal data in the normal database, and when there is a difference value, it indicates that the current device has a fault; continuously collect the operation data of the device to obtain the continuous data of the device, and use a time series graph to visually display the continuous data of the device; obtain the time series analysis result through trend, seasonality, and autocorrelation analysis; input the time series analysis result into the ARIMA model for processing, and optimize the ARIMA model simultaneously; set different prediction steps through the ARIMA model to obtain the future fault prediction result.
[0114] The above-disclosed are only one or more preferred embodiments of the present application, and the scope of rights of the present application cannot be limited thereby. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.
Claims
1. A device for predicting equipment failure, characterized in that: It includes a normal training module, an equipment signal collection module, a current fault prediction module, a time series analysis module and a future fault prediction module; the normal training module, the equipment signal collection module, the current fault prediction module, the time series analysis module and the future fault prediction module are connected in sequence; The normal training module is used to collect historical normal data of the device to obtain a normal database of the device; The device signal collection module is used to collect the current operation data of the device; The current fault prediction module is used to compare the newly collected operation data with the normal data in the normal database, and when a difference value appears, it indicates that the current device has a fault; The time series analysis module is used to continuously record the operation data trajectory of the equipment and perform time series analysis to obtain the time series analysis results; The future fault prediction module is used to perform future fault prediction according to the time series analysis result to obtain a future fault prediction result.
2. The device for predicting equipment failure according to claim 1, characterized in that: The normal training module includes a feature extraction unit and a database construction unit, and the feature extraction unit is connected to the database construction unit; The feature extraction unit is used to query the log and the server, collect the historical normal data, and extract features therefrom to obtain high-value features; The database construction unit is used to store the high-value features into a database, and mark the corresponding fault that occurs when the feature is abnormal on each high-value feature, and finally obtain a normal equipment database.
3. The device for predicting equipment failure according to claim 2, characterized in that: The feature extraction unit comprises a spectrum analysis subunit, a wavelet transformation subunit and a principal component analysis subunit, and the spectrum analysis subunit, the wavelet transformation subunit and the principal component analysis subunit are connected in sequence; The spectrum analysis subunit is used to convert the signal from the time domain to the frequency domain, calculate the spectrum using Fourier transform or fast Fourier transform, and obtain the audio and vibration characteristics of the device; The wavelet transform subunit is used to perform multi-scale decomposition on the signal, extract features using discrete wavelet transform or continuous wavelet transform, and obtain frequency band features of the device signal; The principal component analysis subunit is used to perform dimensionality reduction processing on high-dimensional data, extract main characteristic components or distinguishing features, and package all the extracted features to obtain high-value features.
4. The device for predicting equipment failure according to claim 3, characterized in that: The device signal collection module includes a data collection unit and a data preprocessing unit, and the data collection unit is connected to the data preprocessing unit; The data collection unit is used to collect current operating data through sensors installed on the equipment, and the operating data includes temperature, vibration frequency and pressure; The data preprocessing unit is used to clean and convert the collected operation data.
5. The equipment failure prediction device according to claim 4, characterized in that: The time series analysis module includes a device data continuous recording unit, a trend analysis unit, a seasonal analysis unit and an autocorrelation analysis unit, wherein the device data continuous recording unit, the trend analysis unit, the seasonal analysis unit and the autocorrelation analysis unit are connected in sequence; The device data continuous recording unit is used to continuously collect the operation data of the device to obtain the device continuous data, and use a time series graph to visualize the device continuous data; The trend analysis unit is used to smooth the continuous data of the device using a moving average to identify long-term trends in the data; The seasonal analysis unit is used to extract and analyze the seasonal components of the continuous data of the equipment using seasonal decomposition to obtain seasonal analysis results; The autocorrelation analysis unit is used to calculate the autocorrelation coefficient of the continuous data of the device to understand the correlation data of the data at different time lags, and to package the correlation data, the long-term trend and the seasonal analysis results to obtain a time series analysis result.
6. The equipment failure prediction device according to claim 5, characterized in that: The future fault prediction module includes a data stationarity test unit, an autocorrelation unit, an ARIMA model optimization unit and a future prediction unit, wherein the data stationarity test unit, the autocorrelation unit, the ARIMA model optimization unit and the future prediction unit are connected in sequence; The data stationarity test unit processes the time series analysis result through the ARIMA model and uses the ADF test method to determine whether the sequence is stationary; The autocorrelation unit is used to draw the autocorrelation function and partial autocorrelation function diagrams of the time series data, and preliminarily determine the number of autoregressive terms p and the number of moving average terms q of the ARIMA model by analyzing the ACF and PACF diagrams; The ARIMA model optimization unit is used to optimize, fit and diagnose the ARIMA model; The future prediction unit is used to use the fitted ARIMA model to predict future values, set different prediction steps to obtain prediction values at different time points in the future, and obtain future fault prediction results.
7. The equipment failure prediction device according to claim 6, characterized in that: The ARIMA model optimization unit includes an optimal parameter selection subunit, a model fitting subunit and a model diagnosis subunit, wherein the optimal parameter selection subunit, the model fitting subunit and the model diagnosis subunit are connected in sequence; The optimal parameter selection subunit is used to automatically find the optimal p, d, q parameter combination using the information criterion, and then weigh the complexity and fit of the model to select the best model parameters; The fitting model subunit is used to fit the ARIMA model using the determined parameters p, d, q; The model diagnosis subunit is used to determine whether the residual sequence is a white noise sequence through the ACF and PACF diagrams of the residual sequence and the Ljung-Box Q test.
8. A method for predicting equipment failure, using the equipment failure prediction device according to claim 7, characterized in that: The steps include: By collecting the historical normal data of the equipment, a normal database of the equipment is obtained; Compare the newly collected operation data with the normal data in the normal database, and when a difference value appears, it indicates that the current device has a fault; Continuously collect the operation data of the equipment to obtain continuous data of the equipment, and use a time series chart to visualize the continuous data of the equipment; Obtain time series analysis results through trend, seasonality and autocorrelation analysis; Inputting the time series analysis results into the ARIMA model for processing, and optimizing the ARIMA model at the same time; Different prediction steps are set through the ARIMA model to obtain future fault prediction results.