Method and system for predicting residual life of equipment
By preprocessing and extracting features from equipment data, using a time series prediction model for short-term predictions, and combining this with fault diagnosis model analysis, the problems of low accuracy and credibility caused by non-stationary data in equipment remaining life prediction are solved, achieving more accurate and reliable predictions.
Patent Information
- Application Number
- CN202510775787.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-11
- Publication Date
- 2025-09-12
AI Technical Summary
In the existing technology of equipment remaining life prediction, non-stationary data leads to low prediction accuracy and credibility, and the error accumulation of traditional methods affects the reliability of the prediction results.
By acquiring equipment data in real time, performing data alignment, grouping, and anomaly filtering, and extracting feature data, the system inputs the feature data into a time series prediction model for short-term predictions. The system also uses a fault diagnosis model to analyze the preliminary prediction results, thereby improving prediction accuracy and credibility.
The accuracy and reliability of equipment remaining life prediction are improved, the impact of non-stationary data is reduced, and the accuracy and reliability of prediction are improved.
Smart Images
Figure CN120633438A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of remaining useful life prediction, and in particular to a method and system for predicting the remaining useful life of equipment. Background Art
[0002] A key step in predictive maintenance systems is the prediction of time series data. Time series predictions are generally more accurate when based on stationary data, but are less effective for non-stationary or nearly stationary data. Production data from equipment on a production line often exhibits a certain degree of volatility. During the production process, manual maintenance disrupts the data's changing trends, exacerbating the non-stationarity of production data, which significantly impacts the accuracy of equipment remaining life predictions. Traditional prediction methods, which directly predict a specific point in time, have low accuracy for long timeframes. The gradual accumulation of errors leads to low credibility for future predictions, impacting the practical application of predictive maintenance systems for production line equipment. Summary of the Invention
[0003] In order to solve the above technical problems, the present application provides a method and system for predicting the remaining life of an equipment, which can improve the accuracy and credibility of the remaining life prediction of the equipment.
[0004] Specifically, the present application provides a method for predicting the remaining life of an equipment, comprising the following steps: Acquire original equipment data in real time, and acquire target feature data based on the original equipment data; input the target feature data into a time series prediction model based on the current time series, so as to divide the current time series into several time periods through the time series prediction model, and perform remaining life prediction on the target feature data corresponding to each time period in turn to output a preliminary prediction result; and input the preliminary prediction result into a fault diagnosis model to output a remaining life prediction result.
[0005] In the above technical solution, the remaining life prediction is carried out in time periods to transform long-term predictions into short-term predictions, which can reduce the impact of non-stationary data on the overall model prediction ability. The volatility and trend of each short time segment are more obvious, making the prediction more accurate; by feeding the preliminary prediction results into the fault diagnosis model, the effectiveness and accuracy of the preliminary prediction results can be further analyzed, thereby improving the accuracy and credibility of the final remaining life prediction results.
[0006] Furthermore, the obtaining of target characteristic data includes: preprocessing the original device data based on a first identification code to obtain preliminary device data; wherein the preprocessing includes at least data alignment, data grouping and abnormal data filtering; and extracting features from the preliminary device data based on the first identification code to obtain target characteristic data.
[0007] In the above technical solution, data alignment can ensure that the timestamps of all relevant data are consistent, eliminating time deviations and enabling synchronous analysis of data from different sources. Grouping data according to specific logic (such as device data type) helps to discover the differences and commonalities between different groups, thereby simplifying subsequent data processing and analysis, making it easier for the model to capture key features. By removing abnormal data, these abnormal data can be prevented from having a negative impact on model training, preventing the model from overfitting or underfitting, and avoiding affecting the accuracy of the prediction.
[0008] In addition, by extracting the most useful features for predicting the remaining life of the equipment from the preprocessed data, the data dimension is reduced, the computational complexity is reduced, and at the same time, the information that best reflects the equipment status is retained, thereby improving the predictive ability of the model.
[0009] Furthermore, the time series prediction model includes at least a preprocessing module, a time series grouping module, an encoder module and a prediction module; the time series prediction model performs the following steps: performing secondary preprocessing on the target feature data through the preprocessing module to obtain preliminary prediction data; wherein, the secondary preprocessing includes at least normalization processing and time difference processing; and, dividing the current time series into several time periods through the time series grouping module to obtain preliminary prediction data corresponding to each time period.
[0010] In the above technical solution, the purpose of normalization is to scale the data to a standard range so that data with different features are comparable. Normalized data is usually easier to be processed by optimization algorithms, which accelerates model training. It can also prevent the values of certain features from being too large and dominating the learning process of the model, making the model easier to converge. The purpose of time difference is to eliminate trends in the data by calculating the differences between adjacent data points, making the data smoother, and making it easier for the model to capture short-term fluctuations and anomalies, while reducing the complexity of the model. Short-term predictions are usually more accurate than long-term predictions, reducing error accumulation, and further reducing the complexity of the model.
[0011] Furthermore, the encoder module includes at least a position encoding layer and multiple sequence encoding layers, each sequence encoding layer includes a multi-head self-attention mechanism and a feedforward neural network; the time series prediction model also performs the following steps: sorting the preliminary prediction data corresponding to each time period based on the second identification code through the position encoding layer; and, performing residual connection and layer normalization on the sorted preliminary prediction data based on the multi-head self-attention mechanism and the feedforward neural network through each sequence encoding layer in turn to obtain encoded prediction data.
[0012] In the above technical solution, the position encoding layer enables the model to better understand the temporal dependency and contextual relationship of the data, improve the model's prediction accuracy, and ensure that the preliminary prediction data for each time period is input into the model in the correct time order, which is crucial for capturing the dynamic changes of time series; the multi-head self-attention mechanism can capture the long-term dependencies in the data, enabling the model to understand the correlation between distant time points, improving the model's expressive power, and enabling the model to better capture complex time series features; the feedforward neural network can capture the nonlinear relationships in the data, enabling the model to better adapt to different data distributions.
[0013] In addition, residual connections alleviate the gradient vanishing and exploding problems in deep networks by adding the input directly to the output, making the model easier to train. Layer normalization improves the convergence speed of the model by normalizing the output of each layer, allowing the model to reach the optimal state faster.
[0014] Furthermore, the prediction module includes at least a flattening layer, a fully connected layer and a discarding layer; the time series prediction model also performs the following steps: performing a remaining life prediction on the encoded prediction data through the flattening layer, the fully connected layer and the discarding layer in sequence to obtain preliminary prediction data; and repeating a preset number of remaining life predictions based on the preliminary prediction data through the flattening layer and the fully connected layer to obtain a preliminary prediction result.
[0015] In the above technical solution, the flattening layer simplifies the complex three-dimensional or two-dimensional data structure into one dimension, which facilitates the subsequent fully connected layer to perform global feature processing. While simplifying the data structure, the flattening layer retains the key feature information in the encoded prediction data, ensuring that the subsequent fully connected layer can use this information for prediction; the fully connected layer can capture the global features in the encoded prediction data, enabling the model to understand the characteristics of the data as a whole; the dropout layer reduces the complexity of the model by randomly dropping a part of the neurons, prevents the model from overfitting on the training data, and improves the generalization ability of the model.
[0016] In addition, through multiple predictions, the model can gradually optimize the prediction results, reduce errors, and improve the accuracy of the prediction.
[0017] Furthermore, the fault diagnosis model performs the following steps: obtaining a fault diagnosis sequence based on the preliminary prediction result through the fault diagnosis model, and calculating the remaining life of the equipment according to the fault diagnosis sequence, so as to output a remaining life prediction result based on the calculation result.
[0018] In the above technical solution, a fault diagnosis sequence is generated based on the preliminary prediction results to identify the failure mode or potential failure point of the equipment in different time periods; by combining the fault diagnosis sequence, the remaining life of the equipment can be estimated more accurately.
[0019] Furthermore, in the process of acquiring the original device data, it also includes: storing the target feature data in real time.
[0020] In the above technical solution, real-time storage can ensure data integrity, prevent data loss or damage, and provide reliable data support; real-time storage ensures data consistency, avoids data delays or synchronization issues, and enables models to perform analysis and predictions based on consistent data; real-time stored data is traceable, facilitating data auditing and backtracking, and helping to diagnose and resolve problems.
[0021] Furthermore, the outputting of the remaining life prediction result also includes: outputting warning information based on the remaining life prediction result.
[0022] In the above technical solution, by evaluating the remaining life prediction results and generating warning information, potential faults can be discovered in advance, and preventive measures can be taken to avoid the occurrence of faults.
[0023] Furthermore, based on the same concept, the present application also provides a system for predicting the remaining life of an equipment, the system comprising: The data acquisition module is used to acquire original device data in real time, and acquire target feature data based on the original device data, so as to input the target feature data into the time series prediction module based on the current time series.
[0024] The time series prediction module is used to divide the current time series into several time periods, and perform residual life prediction on the target feature data corresponding to each time period in sequence, so as to output preliminary prediction results to the fault diagnosis module.
[0025] The fault diagnosis module is used to output a remaining life prediction result based on the preliminary prediction result.
[0026] In the above technical solution, long-term predictions are transformed into short-term predictions by dividing time periods, making the predictions more accurate; the remaining life predictions generated based on the preliminary prediction results also further improve the accuracy of the predictions.
[0027] Furthermore, the system also includes: a warning module, which is used to output warning information based on the remaining service life prediction result.
[0028] In the above technical solution, outputting warning information can help users understand the current status of the device, thereby promptly discovering potential failure risks, thereby improving the user experience and device safety.
[0029] Compared with the prior art, the present invention has the following advantages: This application first obtains the original equipment data in real time, and obtains the target feature data based on the original equipment data; then inputs the target feature data into the time series prediction model based on the current time series, so as to divide the current time series into several time periods through the time series prediction model, and performs the remaining life prediction on the target feature data corresponding to each time period in turn to output the preliminary prediction result; and further inputs the preliminary prediction result into the fault diagnosis model to output the remaining life prediction result. This application transforms long-term predictions into short-term predictions, which can reduce the impact of non-stationary data on the overall model prediction ability. The volatility and trend of each short time series segment are more obvious, making the prediction more accurate; by feeding the preliminary prediction results into the fault diagnosis model, the effectiveness and accuracy of the preliminary prediction results can be further analyzed, thereby improving the accuracy and credibility of the final remaining life prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 This is a flow chart of the equipment remaining life prediction method described in this application.
[0031] Figure 2 This is a schematic diagram of time series division in an embodiment of the present application.
[0032] Figure 3 This is a schematic diagram of the operation of the fault diagnosis model described in this application.
[0033] Figure 4 This is a framework diagram of the equipment remaining life prediction system described in this application. DETAILED DESCRIPTION
[0034] The following is a further detailed description of a method and system for predicting the remaining service life of equipment of the present application in conjunction with specific embodiments and drawings.
[0035] For details, see Figure 1 The present application provides a method for predicting the remaining life of equipment, including the following steps S100 to S300. An embodiment is as follows: Taking coating equipment as an example, after obtaining the corresponding solder paste thickness and coating area in real time, time series prediction is performed, fault diagnosis and classification are performed on the predicted sequence, and then the remaining life prediction result of the coating equipment is output.
[0036] Specifically: after obtaining the original equipment data, data preprocessing and feature extraction are first required, and then the extracted target feature data is input into the time series prediction model to realize multivariate prediction of the time series, and output multi-step prediction results as input of the fault diagnosis model; further, the fault diagnosis model is, for example, a binary classification model, which uses the results of the time series prediction as input, identifies the faulty node, and outputs the expected fault processing required after the preset service cycle or preset operating time of the coating equipment, such as cleaning the valve body.
[0037] Steps S100 to S300 are described in detail below.
[0038] Step S100: acquiring original device data in real time, and acquiring target feature data based on the original device data.
[0039] Furthermore, the obtaining of target characteristic data includes: preprocessing the original device data based on a first identification code to obtain preliminary device data; wherein the preprocessing includes at least data alignment, data grouping and abnormal data filtering; and extracting features from the preliminary device data based on the first identification code to obtain target characteristic data.
[0040] In some embodiments, taking the coating equipment as an example, the original equipment data is such as solder paste thickness and coating area; the original equipment data acquired in real time first needs to be aligned, grouped and filtered for abnormal data based on, for example, the product code (i.e., the first identification code); then, key feature data related to the fault (i.e., the target feature data) is extracted based on the product code, such as the maximum value, minimum value, median value, time length and number of changes of the solder paste thickness and coating area.
[0041] It should be noted that in the actual predictive maintenance process, different devices only have different corresponding data input structures, while the remaining links call the same algorithm, and the applied data and model parameters can be configured and expanded by technical personnel in this field according to actual application requirements.
[0042] In the above technical solution, data alignment can ensure that the timestamps of all relevant data are consistent, eliminating time deviations and enabling synchronous analysis of data from different sources. Grouping data according to specific logic (such as device data type) helps to discover the differences and commonalities between different groups, thereby simplifying subsequent data processing and analysis, making it easier for the model to capture key features. By removing abnormal data, these abnormal data can be prevented from having a negative impact on model training, preventing the model from overfitting or underfitting, and avoiding affecting the accuracy of the prediction.
[0043] In addition, by extracting the most useful features for predicting the remaining life of the equipment from the preprocessed data, the data dimension is reduced, the computational complexity is reduced, and at the same time, the information that best reflects the equipment status is retained, thereby improving the predictive ability of the model.
[0044] Furthermore, in the process of acquiring the original device data, it also includes: storing the target feature data in real time.
[0045] In some embodiments, the acquired target feature data is stored and used for subsequent model training, thereby continuously optimizing the model and improving prediction accuracy. At the same time, in addition to being input into the corresponding storage database, the target feature data will also be directly input into the time series prediction model for real-time prediction.
[0046] In the above technical solution, real-time storage can ensure data integrity, prevent data loss or damage, and provide reliable data support; real-time storage ensures data consistency, avoids data delays or synchronization issues, and enables models to perform analysis and predictions based on consistent data; real-time stored data is traceable, facilitating data auditing and backtracking, and helping to diagnose and resolve problems.
[0047] Step S200: Based on the current time series, the target feature data is input into the time series prediction model, so that the current time series is divided into several time periods through the time series prediction model, and the remaining life of the target feature data corresponding to each time period is predicted in turn to output a preliminary prediction result.
[0048] Furthermore, the time series prediction model includes at least a preprocessing module, a time series grouping module, an encoder module and a prediction module; the time series prediction model performs the following steps: performing secondary preprocessing on the target feature data through the preprocessing module to obtain preliminary prediction data; wherein, the secondary preprocessing includes at least normalization processing and time difference processing; and, dividing the current time series into several time periods through the time series grouping module to obtain preliminary prediction data corresponding to each time period.
[0049] In some embodiments, the purpose of normalization is to scale the data to a standard range (such as 0-1 or -1 to 1) to make data with different characteristics comparable; the purpose of time difference is to extract the temporal trend of the data and reduce the noise and fluctuation of the data.
[0050] Furthermore, a Grouping operation is designed to divide the time series into several time periods according to the length and characteristics of the time series, and each time period is a group (e.g. Figure 2As shown in Figure 3, each group is regarded as a token, and a queue consisting of multiple tokens is used as the input of the transformer encoder module. For example, the time period can be divided according to a fixed time interval, and the corresponding preliminary prediction data can be extracted from each time period to form a time period data queue.
[0051] It should be noted that the setting of each group can be overlapping or non-overlapping. Assuming that the total length of the time series is , each group length is , the step size is (non-overlapping area between two consecutive groups), then the number of groups can be expressed as:
[0052] At this time, through group, the number of inputs can be Reduced to approximately .
[0053] In the above technical solutions, normalized data is usually easier to be processed by optimization algorithms, which accelerates model training. It can also prevent the values of certain features from being too large and dominating the model's learning process, making the model easier to converge; time difference makes the data smoother, making it easier for the model to capture short-term fluctuations and anomalies, while reducing the complexity of the model; short-term predictions are usually more accurate than long-term predictions, reducing error accumulation and further reducing the complexity of the model.
[0054] Furthermore, the encoder module includes at least a position encoding layer and multiple sequence encoding layers, each sequence encoding layer includes a multi-head self-attention mechanism and a feedforward neural network; the time series prediction model also performs the following steps: sorting the preliminary prediction data corresponding to each time period based on the second identification code through the position encoding layer; and, performing residual connection and layer normalization on the sorted preliminary prediction data based on the multi-head self-attention mechanism and the feedforward neural network through each sequence encoding layer in turn to obtain encoded prediction data.
[0055] In some embodiments, since the encoder module does not have built-in sequence position information, it is necessary to add position information to each token through position encoding. The position encoding is a learnable vector or a fixed-calculated vector. These encodings are added to the input embedding so that the model can understand the order of tokens.
[0056] The second identification code may be a product code.
[0057] In other embodiments, those skilled in the art may customize the second identification code according to actual application requirements, and the second identification code is not limited to the product code.
[0058] Furthermore, in a real-world Transformer, multiple encoder blocks (i.e., the sequence encoding layer) are chained together to form a deep encoder module. Each encoder block incorporates a multi-head self-attention mechanism and a feedforward neural network, with residual connections and layer normalization applied to each sublayer. This stacked structure enables the model to learn more abstract and high-level feature representations. As the number of layers increases, the model is able to capture longer-range dependencies, improving its ability to understand sequence data.
[0059] In the above technical solution, the position encoding layer enables the model to better understand the temporal dependency and contextual relationship of the data, improve the model's prediction accuracy, and ensure that the preliminary prediction data for each time period is input into the model in the correct time order, which is crucial for capturing the dynamic changes of time series; the multi-head self-attention mechanism can capture the long-term dependencies in the data, enabling the model to understand the correlation between distant time points, improving the model's expressive power, and enabling the model to better capture complex time series features; the feedforward neural network can capture the nonlinear relationships in the data, enabling the model to better adapt to different data distributions.
[0060] In addition, residual connections alleviate the gradient vanishing and exploding problems in deep networks by adding the input directly to the output, making the model easier to train. Layer normalization improves the convergence speed of the model by normalizing the output of each layer, allowing the model to reach the optimal state faster.
[0061] Furthermore, the prediction module includes at least a flattening layer, a fully connected layer and a discarding layer; the time series prediction model also performs the following steps: performing a remaining life prediction on the encoded prediction data through the flattening layer, the fully connected layer and the discarding layer in sequence to obtain preliminary prediction data; and repeating a preset number of remaining life predictions based on the preliminary prediction data through the flattening layer and the fully connected layer to obtain a preliminary prediction result.
[0062] In some embodiments, the output of the Encoder block is flattened into a one-dimensional vector, because the output of the Transformer is usually a three-dimensional tensor, and the fully connected layer generally accepts a one-dimensional vector; the flattened features are nonlinearly transformed using the fully connected layer to extract more meaningful features; further, a portion of neurons are randomly discarded through the dropout layer, and then the combination of the flattening layer and the fully connected layer can be repeated as needed, and finally a preliminary prediction result is obtained through the output layer.
[0063] In the above technical solution, the flattening layer simplifies the complex three-dimensional or two-dimensional data structure into one dimension, which facilitates the subsequent fully connected layer to perform global feature processing. While simplifying the data structure, the flattening layer retains the key feature information in the encoded prediction data, ensuring that the subsequent fully connected layer can use this information for prediction; the fully connected layer can capture the global features in the encoded prediction data, enabling the model to understand the characteristics of the data as a whole; the dropout layer can reduce the complexity of the model, prevent the model from overfitting on the training data, and improve the generalization ability of the model.
[0064] In addition, through multiple predictions, the model can gradually optimize the prediction results, reduce errors, and improve the accuracy of the prediction.
[0065] Step S300: inputting the preliminary prediction result into a fault diagnosis model to output a remaining life prediction result.
[0066] Furthermore, the fault diagnosis model performs the following steps: obtaining a fault diagnosis sequence based on the preliminary prediction result through the fault diagnosis model, and calculating the remaining life of the equipment according to the fault diagnosis sequence, so as to output a remaining life prediction result based on the calculation result.
[0067] In some embodiments, fault diagnosis uses the existing Hydra-MultiRocket time series classification model to classify the predicted multivariate sequence and obtain a classification result, which is a vector. Figure 3 As shown, the preliminary prediction results are Figure 3 The time series prediction sequence in is input into the diagnosis model to obtain a fault diagnosis sequence.
[0068] Furthermore, the impact of each fault on the equipment life is evaluated based on the fault modes and their occurrence events in the fault diagnosis sequence. For example, overload may accelerate equipment wear.
[0069] The remaining life of the equipment can be calculated based on historical data and the influencing factors corresponding to the failure mode, and then the calculated remaining life result can be formatted and output, such as the remaining operating time and remaining service life of the equipment. For example, the output can be "the remaining life of the equipment is 1000 hours of operation" or "the remaining life of the equipment is 100 times of use."
[0070] In the above technical solution, a fault diagnosis sequence is generated based on the preliminary prediction results to identify the failure mode or potential failure point of the equipment in different time periods; by combining the fault diagnosis sequence, the remaining life of the equipment can be estimated more accurately.
[0071] Furthermore, the outputting of the remaining life prediction result also includes: outputting warning information based on the remaining life prediction result.
[0072] In some embodiments, a corresponding warning message is generated based on the remaining life prediction result. For example, if the remaining life is less than 100 hours, a warning message is generated: "The equipment is about to fail. Please arrange maintenance as soon as possible." Taking the coating equipment as an example, the maintenance may be cleaning the valve body.
[0073] In other embodiments, graded warnings can also be set, and different warning levels can be set according to different ranges of remaining life; for example, if the remaining life is greater than 200 hours, a warning message of "normal operation, no special attention required" is generated; if the remaining life is greater than 100 hours but less than or equal to 200 hours, a warning message of "warning, maintenance recommended" is generated; if the remaining life is less than or equal to 100 hours, a warning message of "serious warning, immediate maintenance" is generated.
[0074] In addition, the warning information can be pushed to relevant management personnel through the interactive interface of the host computer, can be prompted synchronously through voice, and corresponding warning lights can be set on the equipment for warning.
[0075] In the above technical solution, by evaluating the remaining life prediction results and generating warning information, potential faults can be discovered in advance, and preventive measures can be taken to avoid the occurrence of faults.
[0076] Further, based on the same concept, see Figure 4 , the present application also provides a system for predicting the remaining life of an equipment, the system comprising: The data acquisition module is used to acquire original device data in real time, and acquire target feature data based on the original device data, so as to input the target feature data into the time series prediction module based on the current time series.
[0077] The time series prediction module is used to divide the current time series into several time periods, and perform residual life prediction on the target feature data corresponding to each time period in sequence, so as to output preliminary prediction results to the fault diagnosis module.
[0078] The fault diagnosis module is used to output a remaining life prediction result based on the preliminary prediction result.
[0079] In the above technical solution, long-term predictions are transformed into short-term predictions by dividing time periods, making the predictions more accurate; the remaining life predictions generated based on the preliminary prediction results also further improve the accuracy of the predictions.
[0080] Furthermore, the system also includes: a warning module, which is used to output warning information based on the remaining service life prediction result.
[0081] In the above technical solution, outputting warning information can help users understand the current status of the device, thereby promptly discovering potential failure risks, thereby improving the user experience and device safety.
[0082] It should be noted that the equipment remaining life prediction system and the equipment remaining life prediction method are based on the same inventive concept, so the embodiments of the system part are the same as the various steps of the above method and will not be repeated here.
[0083] In summary, the present application provides a method and system for predicting the remaining life of an equipment; first, the original equipment data is acquired in real time, and target feature data is acquired based on the original equipment data; then, the target feature data is input into a time series prediction model based on the current time series, so that the current time series is divided into several time periods through the time series prediction model, and the target feature data corresponding to each time period is predicted in turn to output a preliminary prediction result; and the preliminary prediction result is further input into a fault diagnosis model to output a remaining life prediction result. The present application transforms long-term predictions into short-term predictions, which can reduce the impact of non-stationary data on the overall model prediction ability, and the volatility and trend of each short time series segment are more obvious, making the prediction more accurate; by feeding the preliminary prediction results into the fault diagnosis model, the effectiveness and accuracy of the preliminary prediction results can be further analyzed, thereby improving the accuracy and credibility of the final remaining life prediction results.
[0084] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely illustrative and are not intended to limit the scope of the present application. Various changes and modifications may be made therein by those skilled in the art without departing from the scope and spirit of the present application. All such changes and modifications are intended to be included within the scope of the present application as required by the appended claims.
[0085] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0086] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units described is merely a logical function division. In actual implementation, other division methods may be used, such as combining or integrating multiple units or components into another device, or ignoring or not performing some features.
[0087] The various component embodiments of the present application can be implemented in hardware, or in a software module running on one or more processors, or in a combination thereof. Those skilled in the art will appreciate that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some modules according to the embodiments of the present application. The application can also be implemented as a part or all of a device program (e.g., a computer program and a computer program product) for performing the method described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.
[0088] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.
[0089] Although the present application is described in conjunction with the above specific embodiments, it is obvious that those skilled in the art can make many substitutions, modifications and variations based on the above content. Therefore, all such substitutions, improvements and variations are included in the spirit and scope of the appended claims.
Claims
1. A method for predicting the remaining life of equipment, characterized in that: The following steps are involved: Acquire raw device data in real time, and acquire target feature data based on the raw device data; Inputting the target feature data into a time series prediction model based on the current time series, so as to divide the current time series into a number of time periods through the time series prediction model, and sequentially performing remaining life prediction on the target feature data corresponding to each time period, so as to output a preliminary prediction result; And, the preliminary prediction result is input into a fault diagnosis model to output a remaining life prediction result.
2. The method for predicting the remaining life of equipment according to claim 1, characterized in that: The acquiring target feature data includes: Preprocessing the original device data based on the first identification code to obtain preliminary device data; wherein the preprocessing includes at least data alignment, data grouping, and abnormal data filtering; And, feature extraction is performed on the preparation device data based on the first identification code to obtain target feature data.
3. The method for predicting the remaining life of equipment according to claim 2, characterized in that: The time series prediction model includes at least a preprocessing module, a time series grouping module, an encoder module, and a prediction module; the time series prediction model performs the following steps: Performing secondary preprocessing on the target feature data by the preprocessing module to obtain preliminary prediction data; wherein the secondary preprocessing includes at least normalization processing and time difference processing; Furthermore, the current time series is divided into several time periods by the time series grouping module to obtain preliminary prediction data corresponding to each time period.
4. The method for predicting the remaining life of equipment according to claim 3, characterized in that: The encoder module includes at least a position encoding layer and multiple sequence encoding layers, each sequence encoding layer includes a multi-head self-attention mechanism and a feedforward neural network; the time series prediction model further performs the following steps: sorting the prepared prediction data corresponding to each time period based on the second identification code through the position coding layer; Furthermore, residual connection and layer normalization are performed on the sorted preliminary prediction data in turn based on the multi-head self-attention mechanism and the feedforward neural network through each sequence encoding layer to obtain the encoded prediction data.
5. The method for predicting the remaining life of equipment according to claim 4, characterized in that: The prediction module includes at least a flattening layer, a fully connected layer, and a discard layer; the time series prediction model further performs the following steps: Performing a remaining life prediction on the coded prediction data through the flattening layer, the fully connected layer, and the discard layer in sequence to obtain preliminary prediction data; Furthermore, the preset remaining life prediction is repeatedly performed based on the preliminary prediction data through the flattening layer and the fully connected layer to obtain a preliminary prediction result.
6. The method for predicting the remaining life of equipment according to claim 5, characterized in that: The fault diagnosis model performs the following steps: A fault diagnosis sequence is obtained based on the preliminary prediction result through the fault diagnosis model, and the remaining life of the equipment is calculated according to the fault diagnosis sequence, so as to output a remaining life prediction result based on the calculation result.
7. The method for predicting the remaining life of equipment according to claim 1, characterized in that: The process of acquiring the original device data also includes: storing the target feature data in real time.
8. The method for predicting the remaining life of equipment according to claim 6, characterized in that: The outputting of the remaining life prediction result further includes: outputting warning information based on the remaining life prediction result.
9. A system using the equipment remaining life prediction method according to any one of claims 1 to 8, characterized in that: The system comprises: A data acquisition module, configured to acquire raw device data in real time, and acquire target feature data based on the raw device data, so as to input the target feature data into a time series prediction module based on a current time series; The time series prediction module is used to divide the current time series into several time periods and perform residual life prediction on the target feature data corresponding to each time period in sequence, so as to output preliminary prediction results to the fault diagnosis module; The fault diagnosis module is used to output a remaining life prediction result based on the preliminary prediction result.
10. The system according to claim 9, characterized in that The system further comprises: The warning module is used to output warning information based on the remaining service life prediction result.