Equipment fault prediction method, prediction device and electronic equipment

Abstract

The invention relates to the technical field of fault prediction, and provides an equipment fault prediction method and device and electronic equipment. According to the method, the equipment operation data is acquired, the fault prediction model is constructed based on the equipment operation data, and the constructed fault prediction model is used to predict the real-time operation data of the equipment at the second moment to obtain the fault prediction result. And when it is determined that the target index of the fault prediction model is smaller than a preset index threshold value based on the fault prediction result and the rechecking mark information of the fault prediction result, the equipment operation data is updated based on the real-time operation data at the second moment, and then the fault prediction model is updated based on the updated equipment operation data. And finally, the updated fault prediction model is used for predicting the equipment fault, the real-time data of equipment operation can be analyzed and processed, the fault prediction model is constructed in combination with an incremental learning method, model parameters are updated in real time to improve prediction accuracy and real-time performance, the maintenance cost is reduced, and the equipment safety is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault prediction, and in particular to a device fault prediction method, a prediction device and an electronic device. BACKGROUND

[0002] In the industrial production industry, stable operation of equipment is crucial for production efficiency and safety. With the complication of production equipment and processes, traditional fault prediction methods face the following challenges:

[0003] 1) Large amount of data: the amount of data in industrial production processes is huge, including process parameters, equipment status, operation data, vibration data, video data, etc., which need to be processed and analyzed in real time.

[0004] 2) Changing working conditions: production conditions and equipment status often change, and traditional static models are difficult to adapt to dynamic changes in working conditions, affecting prediction accuracy.

[0005] 3) Prospective requirement: fault prediction needs to give timely warning before equipment failure occurs, so that preventive measures can be taken. Traditional methods only make judgments based on past experience values on current running data, which has defects in the foresight of prediction.

[0006] Therefore, a new device fault prediction method needs to be proposed to improve prediction accuracy and real-time performance. SUMMARY

[0007] Therefore, the embodiments of the present application provide a device fault prediction method, a prediction device and an electronic device to solve the problem that the device fault prediction algorithm in the prior art is not real-time and cannot adapt to complex working conditions.

[0008] The first aspect of the embodiments of the present application provides a device fault prediction method, comprising:

[0009] Obtaining device operation data, the device operation data including at least one of real-time operation data of the device at a first time and historical operation data of the device;

[0010] Constructing a fault prediction model based on the device operation data, the fault prediction model including at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer and an output layer; wherein the input layer is used to input the device operation data; the convolutional neural network layer is used to extract features from the device operation data to obtain initial feature data; the gated recurrent unit and the attention layer are used to capture long-term dependencies and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is used to obtain a fault prediction result based on the target feature data;

[0011] The acquisition device acquires real-time operation data of the device at a second time, inputs the real-time operation data at the second time into the fault prediction model, and obtains a fault prediction result; the second time is later than the first time;

[0012] The review mark information of the fault prediction result is acquired, and in response to determining that a target index of the fault prediction model based on the fault prediction result and the review mark information is less than a preset index threshold, the device operation data is updated based on the real-time operation data at the second time, and the fault prediction model is updated based on the updated device operation data; the target index includes at least one of a fault prediction accuracy and a fault prediction score;

[0013] The updated fault prediction model is used to predict a fault of the device.

[0014] In a second aspect of the embodiments of the present application, a device fault prediction apparatus is provided, including:

[0015] An acquisition module is configured to acquire device operation data, the device operation data including at least one of real-time operation data of the device at a first time and historical operation data of the device;

[0016] A construction module is configured to construct a fault prediction model based on the device operation data, the fault prediction model including at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer, and an output layer; the input layer is used to input the device operation data; the convolutional neural network layer is used to perform feature extraction on the device operation data to obtain initial feature data; the gated recurrent unit and the attention layer are used to capture long-term dependency relationships and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is used to obtain a fault prediction result based on the target feature data;

[0017] A prediction module is configured to acquire real-time operation data of the device at a second time, input the real-time operation data at the second time into the fault prediction model, and obtain a fault prediction result; the second time is later than the first time;

[0018] An update module is configured to acquire review mark information of the fault prediction result, and in response to determining that a target index of the fault prediction model based on the fault prediction result and the review mark information is less than a preset index threshold, update the device operation data based on the real-time operation data at the second time, and update the fault prediction model based on the updated device operation data; the target index includes at least one of a fault prediction accuracy and a fault prediction score;

[0019] The prediction module is further configured to use the updated fault prediction model to predict a fault of the device.

[0020] In a third aspect, the embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method when executing the computer program.

[0021] Compared with the prior art, the embodiment of the present application has the beneficial effects that: the embodiment of the present application obtains the equipment running data, constructs the fault prediction model based on the equipment running data, uses the constructed fault prediction model to predict the real-time running data of the equipment at the second time, obtains the fault prediction result, and when the target index of the fault prediction model is less than the preset index threshold based on the fault prediction result and the review mark information of the fault prediction result, updates the equipment running data based on the real-time running data at the second time, updates the fault prediction model based on the updated equipment running data, and finally uses the updated fault prediction model to predict the equipment fault, which can analyze and process the real-time data of the equipment running, and construct the fault prediction model combined with the incremental learning method, update the model parameters in real time to improve the prediction accuracy and real-time performance, reduce the maintenance cost and improve the equipment safety. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Figure 1 is a flowchart of a device fault prediction method provided by the embodiment of the present application.

[0024] Figure 2 is a flowchart of a method for determining the convolution kernel size and the convolution layer number of a convolutional neural network layer provided by the embodiment of the present application.

[0025] Figure 3 is a flowchart of a method for determining the hidden unit number and the attention mechanism head number of a gated recurrent unit provided by the embodiment of the present application.

[0026] Figure 4 is a flowchart of a method for determining the regularization factor of an attention layer provided by the embodiment of the present application.

[0027] Figure 5 is a flowchart of a method for determining the fault prediction accuracy provided by the embodiment of the present application.

[0028] Figure 6 is a flowchart of a method for determining the fault prediction score provided by the embodiment of the present application.

[0029] Figure 7 FIG. 1 is a flowchart of a method for updating equipment operation data based on real-time operation data at a second time point according to an embodiment of the present application.

[0030] Figure 8 FIG. 2 is a flowchart of a method for updating a fault prediction model based on updated equipment operation data according to an embodiment of the present application.

[0031] Figure 9 FIG. 3 is a flowchart of another method for predicting equipment faults according to an embodiment of the present application.

[0032] Figure 10 FIG. 4 is a schematic diagram of an equipment fault prediction device according to an embodiment of the present application.

[0033] Figure 11 FIG. 5 is a schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0034] In the following description, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the present application. However, persons skilled in the art will understand that the present application can be practiced without such specific details. In other instances, well-known structures, devices, circuits, and methods have not been described in detail in order to avoid obscuring the present application.

[0035] A method and device for predicting equipment faults according to an embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0036] As mentioned above, the traditional fault prediction method performs poorly when facing a large amount of data and complex working conditions, and cannot achieve forward-looking early warning. In the related art, some solutions use incremental learning algorithms such as support vector machines and decision trees to perform real-time analysis and fault prediction on equipment state data. This type of method can implement a gradual learning and updating mechanism, but may have limitations when processing complex working conditions and high-dimensional data.

[0037] In view of this, the embodiment of the present application provides a device fault prediction method, which obtains device operation data, constructs a fault prediction model based on the device operation data, uses the constructed fault prediction model to predict the real-time operation data of the device at a second time, obtains a fault prediction result, and then determines that the target index of the fault prediction model is less than a preset index threshold based on the fault prediction result and review mark information of the fault prediction result, updates the device operation data based on the real-time operation data at the second time, updates the fault prediction model based on the updated device operation data, and finally uses the updated fault prediction model to predict device faults. The real-time data of the device operation can be analyzed and processed, and the fault prediction model is constructed by combining the incremental learning method, the model parameters are updated in real time to improve the prediction accuracy and real-time performance, the maintenance cost is reduced, and the device safety is improved.

[0038] Figure 1 is a flowchart of a device fault prediction method provided by the embodiment of the present application. As shown in Figure 1 , the method comprises the following steps:

[0039] In step S101, device operation data is obtained.

[0040] The device operation data comprises at least one of real-time operation data of the device at a first time and historical operation data of the device.

[0041] In step S102, a fault prediction model is constructed based on the device operation data.

[0042] The fault prediction model comprises at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer and an output layer; the input layer is used to input the device operation data; the convolutional neural network layer is used to extract features of the device operation data to obtain initial feature data; the gated recurrent unit and the attention layer are used to capture long-term dependencies and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is used to obtain a fault prediction result based on the target feature data.

[0043] In step S103, real-time operation data of the device at a second time is obtained, the real-time operation data at the second time is input into the fault prediction model, and a fault prediction result is obtained.

[0044] The second time is later than the first time.

[0045] In step S104, review mark information of the fault prediction result is obtained, and in response to determining that the target index of the fault prediction model is less than a preset index threshold based on the fault prediction result and the review mark information, the device operation data is updated based on the real-time operation data at the second time, and the fault prediction model is updated based on the updated device operation data.

[0046] The target index includes at least one of a fault prediction accuracy and a fault prediction score.

[0047] In step S105, the updated fault prediction model is used to predict the fault of the device.

[0048] In some embodiments of the present application, the method can be performed by a server or a terminal device with certain computing power.

[0049] In some embodiments of the present application, the device operation data can be obtained first, which can include at least one of real-time operation data of the device at a first time and historical operation data of the device.

[0050] The real-time operation data of the device at the first time can be real-time data obtained from various sensors connected to the device, such as real-time data obtained from a sensing and control device that collects data collected by various sensors, such as temperature sensor data, pressure sensor data, flow meter data, and vibration sensor data. The first time can be the current time when the fault prediction model is constructed. Meanwhile, the device historical data. The historical operation data of the device can be obtained from the device control system or the historical record database.

[0051] In some embodiments of the present application, a fault prediction model can be constructed based on the device operation data. The fault prediction model includes at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer, and an output layer. The fault prediction model can be constructed based on the device operation data by configuring parameters of each layer or unit in the fault prediction model.

[0052] In some embodiments of the present application, the input layer is used to input the device operation data; the convolutional neural network layer is used to extract features from the device operation data to obtain initial feature data; the gated recurrent unit and the attention layer are used to capture long-term dependencies and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is used to obtain a fault prediction result based on the target feature data. The parameters of at least part of the layers or units in the input layer, the convolutional neural network layer, the gated recurrent unit, the attention layer, and the output layer can be configured based on the device operation data, thereby obtaining the constructed fault prediction model.

[0053] In some embodiments of the present application, real-time operation data of the device at a second time point later than the first time point can also be obtained. For example, the real-time operation data at the first time point can be the real-time operation data of the device when the fault prediction model is constructed, and the real-time operation data at the second time point can be the real-time operation data of the device when the constructed fault prediction model is used to predict the operation fault of the device. The real-time operation data at the second time point is input into the constructed fault prediction model, and a fault prediction result can be obtained.

[0054] The fault prediction result can be reviewed to obtain review mark information. In an example, a target index of the fault prediction model can also be determined based on the fault prediction result and the review mark information, and the target index includes at least one of a fault prediction accuracy and a fault prediction score. If it is determined that the target index is greater than or equal to a preset index threshold, the existing fault prediction model can be continued to be used without updating. Otherwise, if it is determined that the target index is less than the preset index threshold, the device operation data can be updated based on the real-time operation data at the second time point, and the fault prediction model can be updated based on the updated device operation data. Then, the updated fault prediction model can be used to predict the fault of the device.

[0055] According to the technical scheme provided by the embodiments of the present application, by obtaining device operation data, constructing a fault prediction model based on the device operation data, using the constructed fault prediction model to predict the real-time operation data of the device at a second time point, obtaining a fault prediction result, and when it is determined that a target index of the fault prediction model based on the fault prediction result and review mark information of the fault prediction result is less than a preset index threshold, updating the device operation data based on the real-time operation data at the second time point, updating the fault prediction model based on the updated device operation data, and finally using the updated fault prediction model to predict the fault of the device, the real-time data of the device operation can be analyzed and processed, and the fault prediction model can be constructed by combining the incremental learning method, the model parameters can be updated in real time to improve the prediction accuracy and real-time performance, and the maintenance cost can be reduced and the safety of the device can be improved.

[0056] In some embodiments of the present application, the device operation data can be pre-processed data, and the pre-processing can include cleaning and normalization processing.

[0057] The data collected in real time or obtained from the database can be referred to as raw data, and the format of the raw data is usually as shown in Table 1:

[0058] Table 1 Raw data format

[0059]

[0060]

[0061] The original data can be cleaned and converted to ensure its quality and consistency, laying a foundation for subsequent analysis and modeling. The data cleaning process includes removing redundant records and detecting outliers; data conversion uses normalization to scale the data to a specific range to eliminate the influence of dimension.

[0062] The cleaned data is shown in Table 2.

[0063] Table 2 Cleaned data

[0064] ID Temperature (°C) Pressure (Pa) Flow (L / min) …… Fault status Fault type 1 85 101325 150 …… 0 No fault 2 90 101400 145 …… 0 No fault 3 88 101500 160 …… 1 Pump fault 4 87 01300 155 …… 0 No fault 5 92 1016001 158 …… 1 Sensor fault …… …… …… …… …… …… ……

[0065] The embodiment of the present application can use the min-max normalization processing method to normalize the cleaned data. The working principle of min-max normalization is to map the data to the interval between 0 and 1 by subtracting the minimum value and dividing by the difference between the maximum value and the minimum value. The normalized data is shown in Table 3.

[0066] Table 3 Normalized data

[0067]

[0068]

[0069] The final data format of the equipment operation data can be:

[0070]

[0071] In some embodiments, the preprocessed equipment operation data can be two-dimensionally unfolded and reorganized in the input layer. In an example, if the preprocessed equipment operation data is represented as where x i is the input feature, y i is the fault type, n is the feature dimension, and N represents the number of samples, the input layer can be used to expand each 1x n data to n x n. Taking data x i as an example, the expansion process can be:

[0072]

[0073] In some embodiments of the present application, the parameters of the convolutional neural network layer can include the convolution kernel size and the number of convolution layers.

[0074] Figure 2 is a flowchart of a method for determining the convolution kernel size and the number of convolution layers of a convolutional neural network layer provided by the embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0075] In step S201, the data type of the device running data is acquired.

[0076] The data type includes high-frequency sensor data and low-frequency sensor data, the high-frequency sensor data is data with a sampling frequency greater than or equal to a preset frequency threshold, and the low-frequency sensor data is data with a sampling frequency less than the preset frequency threshold.

[0077] In step S202, in response to determining that the data type is high-frequency sensor data, the convolution kernel size of the convolutional neural network layer is set to a first size, and the number of convolutional layers is set to a first number.

[0078] In step S203, in response to determining that the data type is low-frequency sensor data, the convolution kernel size of the convolutional neural network layer is set to a second size, and the number of convolutional layers is set to a second number.

[0079] The first size is less than the second size, and the first number is greater than the second number.

[0080] In some embodiments of the present application, the convolution kernel size and the number of convolutional layers of the convolutional neural network layer can be set to different values according to different data types. The data type can include high-frequency sensor data and low-frequency sensor data, the high-frequency sensor data is data with a sampling frequency greater than or equal to a preset frequency threshold, and the low-frequency sensor data is data with a sampling frequency less than the preset frequency threshold. The value of the preset frequency threshold can be set according to actual needs, and in an example, it can be set to 1 kHz (kilohertz).

[0081] If the data type of the device running data is high-frequency sensor data, the convolution kernel size of the convolutional neural network layer can be set to a first size, and the number of convolutional layers can be set to a first number. On the other hand, if the data type of the device running data is low-frequency sensor data, the convolution kernel size of the convolutional neural network layer can be set to a second size, and the number of convolutional layers can be set to a second number. The first size is less than the second size, and the first number is greater than the second number.

[0082] In an example, the first size can be set to 3x3, and the first number can be set to 3 layers, so that the set convolutional neural network layer can capture fine-grained features in the device running data; at the same time, the second size can be set to 5x5, and the second number can be set to 2 layers, so that the set convolutional neural network layer can extract global trends.

[0083] In some embodiments of the present application, the parameters of the gated recurrent unit can include the number of hidden units and the number of attention mechanism heads.

[0084] Figure 3 is a flowchart of a method for determining the number of hidden units and the number of attention mechanism heads of a gated recurrent unit provided by an embodiment of the present application. As shown inFigure 3 As shown, the method includes the following steps:

[0085] In step S301, the operating condition fluctuation coefficient of the equipment is determined based on the standard deviation of the equipment operating data.

[0086] In step S302, in response to determining that the operating condition fluctuation coefficient is greater than or equal to a preset fluctuation threshold, the number of hidden units in the gated loop unit is set to a first value, and the number of attention mechanism heads is set to a second value.

[0087] In step S303, in response to determining that the operating condition fluctuation coefficient is less than the preset fluctuation threshold, the number of hidden units of the gated loop unit is set to the third value, and the number of attention mechanism heads is set to the fourth value.

[0088] Among them, the first value is greater than the third value, and the second value is greater than the fourth value.

[0089] In some embodiments of this application, the number of hidden units and the number of attention mechanism heads in the gated loop unit can be set to different values ​​according to the different operating condition fluctuation coefficients of the device operating data. The operating condition fluctuation coefficient of the device operating data can be determined based on the standard deviation of the device operating data. If the operating condition fluctuation coefficient is greater than or equal to a preset fluctuation threshold, the current operating condition can be considered a transient condition; conversely, if the operating condition fluctuation coefficient is less than the preset fluctuation threshold, the current operating condition can be considered a steady-state condition. The value of the preset fluctuation threshold can be determined according to actual needs. In one example, the preset fluctuation threshold can be set to 0.5.

[0090] If the operating condition fluctuation coefficient is greater than or equal to the preset fluctuation threshold, the number of hidden units in the gated loop unit can be set to the first value, and the number of attention mechanism heads can be set to the second value. Conversely, if the operating condition fluctuation coefficient is less than the preset fluctuation threshold, the number of hidden units in the gated loop unit can be set to the third value, and the number of attention mechanism heads can be set to the fourth value. The first value is greater than the third value, and the second value is greater than the fourth value.

[0091] In one example, the first value can be set to 128, and the second value to 4. That is, for equipment operation data collected under transient conditions, the number of hidden units in the gated loop unit can be set to 128, and the attention mechanism can be set to 4 heads. On the other hand, the third value can be set to 64, and the fourth value to 1. That is, for equipment operation data collected under stable conditions, the number of hidden units in the gated loop unit can be set to 64, and the attention mechanism can be set to single head.

[0092] In some embodiments of this application, the attention layer can determine the correlation weight between each time step in the initial feature data and the current prediction time based on a regularization factor. In other words, the parameters of the attention layer may include a regularization factor.

[0093] Figure 4 is a flowchart of a method for determining a regularization factor of an attention layer provided by an embodiment of the present application. As shown in the figure, the method comprises the following steps: Figure 4

[0094] In step S401, the number of sensors and the number of fault types corresponding to the equipment operation data are obtained.

[0095] In step S402, the ten logarithm value of the sum of the number of sensors and the number of fault types is multiplied by a preset weight coefficient to obtain a regularization factor.

[0096] In some embodiments of the present application, the regularization factor of the attention layer can be obtained by first obtaining the number of sensors and the number of fault types corresponding to the equipment operation data, and then multiplying the ten logarithm value of the sum of the number of sensors and the number of fault types by a preset weight coefficient.

[0097] In an example, if the number of sensors is denoted as N sensors and the number of fault types is denoted as N fault_types , then the regularization factor λ is: λ = a·log(N sensors +N fault_types ), where the log function is based on 10, a is the regularization factor, and the value of a can be 0.1, for example.

[0098] After passing through the input layer, the convolutional neural network layer, the gated recurrent unit, and the attention layer, a feature H can be obtained. In the output layer, the feature H can be connected through a fully connected layer to obtain a feature z = WH + b, where W and b are both parameters to be learned. Then the feature z can be converted into a probability distribution through a softmax (normalized exponential) function, and a fault prediction result is obtained. The conversion process can be where k is the number of data contained in the feature z.

[0099] In this way, the fault prediction model can be automatically constructed based on the equipment operation data, and the constructed fault prediction model can be used to continue predicting faults based on real-time data of the equipment operation to obtain a fault prediction result.

[0100] In some embodiments of the present application, the fault prediction result predicted by the fault prediction model can be reviewed, and the review result can be marked to obtain review marking information. The review can be implemented in an artificial review manner, or can be implemented in other manners, which is not limited here.

[0101] ​The target index of the fault prediction model can be obtained by comparing the fault prediction result and the review label information, and the target index includes at least one of a fault prediction accuracy and a fault prediction score.

[0102] Figure 5 is a flowchart of a method for determining a fault prediction accuracy provided by an embodiment of the present application. As shown in Figure 5 , the method includes the following steps:

[0103] In step S501, a number of correctly predicted positive categories, a number of correctly predicted negative categories, a number of incorrectly predicted positive categories, and a number of incorrectly predicted negative categories are determined based on the fault prediction result and the review label information.

[0104] In step S502, a first sum value is determined as a sum of the number of correctly predicted positive categories and the number of correctly predicted negative categories.

[0105] In step S503, a second sum value is determined as a sum of the number of correctly predicted positive categories, the number of correctly predicted negative categories, the number of incorrectly predicted positive categories, and the number of incorrectly predicted negative categories.

[0106] In step S504, a fault prediction accuracy is determined as a quotient of the first sum value and the second sum value.

[0107] In some embodiments of the present application, when determining the fault prediction accuracy, the number of correctly predicted positive categories TP, the number of correctly predicted negative categories TN, the number of incorrectly predicted positive categories FP, and the number of incorrectly predicted negative categories FN can be first determined based on the fault prediction result and the review label information, then the first sum value is determined as a sum of the number of correctly predicted positive categories and the number of correctly predicted negative categories, the second sum value is determined as a sum of the number of correctly predicted positive categories, the number of correctly predicted negative categories, the number of incorrectly predicted positive categories, and the number of incorrectly predicted negative categories, and finally the fault prediction accuracy is determined as a quotient of the first sum value and the second sum value.

[0108] That is, the fault prediction accuracy Accuracy can be calculated by the following formula:

[0109]

[0110] Figure 6 is a flowchart of a method for determining a fault prediction score provided by an embodiment of the present application. As shown in Figure 6 , the method includes the following steps:

[0111] In step S601, a prediction precision and a prediction recall are determined based on the fault prediction result and the review label information.

[0112] In step S602, twice the product of the prediction accuracy rate and the prediction recall rate is determined as a first product.

[0113] In step S603, the sum of the prediction accuracy rate and the prediction recall rate is determined as a third sum value.

[0114] In step S604, the quotient of the first product and the third sum value is determined as the failure prediction score.

[0115] In some embodiments of the present application, in determining the failure prediction score, the prediction accuracy rate P and the prediction recall rate R can be first determined based on the failure prediction result and the review label information, then twice the product of the prediction accuracy rate and the prediction recall rate is determined as the first product, the sum of the prediction accuracy rate and the prediction recall rate is determined as the third sum value, and finally the quotient of the first product and the third sum value is determined as the failure prediction score.

[0116] That is, the failure prediction score F1 can be calculated by the following formula:

[0117] After the target indicators such as the failure prediction accuracy rate and the failure prediction score are determined, the target indicators can be compared with preset indicator thresholds. If the target indicators are greater than or equal to the preset indicator thresholds, the failure prediction model can not be updated. Otherwise, if the target indicators are less than the preset indicator thresholds, the failure prediction model can be updated using the device running real-time data, that is, the failure prediction model is subjected to incremental learning to improve the scene adaptability of the failure prediction model.

[0118] The preset indicator thresholds can be set according to actual needs, for example, the historical failure data and the KPI (Key Performance Indicator) value of the failure prediction model can be combined, and the practice and standards of similar application scenarios in the industry are also referred to. In an example, Accuracy can be set to 85%, and F1 can be set to 80. The preset indicator thresholds can also be adjusted according to actual business needs, which are not limited here.

[0119] In updating the failure prediction model using the device running real-time data, the device running data can be first updated based on the real-time running data at the second time, and then the failure prediction model is updated based on the updated device running data.

[0120] Figure 7 is a flowchart of a method for updating device running data based on real-time running data at a second time provided by an embodiment of the present application. As shown in Figure 7 the method includes the following steps:

[0121] In step S701, a first minimum data in the device running data and a second minimum data in real-time running data at a second time, and a first maximum data in the device running data and a second maximum data in real-time running data at the second time are respectively acquired.

[0122] The first minimum data and the first maximum data are normalized data, and the second minimum data and the second maximum data are original data.

[0123] In step S702, a sum of a product of the first coefficient and the first minimum data and a product of the second coefficient and the second minimum data is determined as an updated normalized interval minimum data value.

[0124] In step S703, a sum of a product of the first coefficient and the first maximum data and a product of the second coefficient and the second maximum data is determined as an updated normalized interval maximum data value.

[0125] In step S704, based on the updated normalized interval minimum data value and the updated normalized interval maximum data value, an updated normalized interval is determined, and sample data is mapped to the updated normalized interval to update the device running data.

[0126] The sample data includes real-time running data at the second time, or includes the device running data and the real-time running data at the second time; and the first coefficient is greater than the second coefficient.

[0127] In some embodiments of the present application, updating the device running data based on the real-time running data at the second time can be that first a first minimum data x min t in the device running data and a second minimum data min(t+1) in real-time running data at a second time are acquired, and then a first maximum data x max t in the device running data and a second maximum data max(t+1) in real-time running data at the second time are acquired. Wherein, x min t is a minimum data in the device running data after normalization processing, x max t is a maximum data in the device running data after normalization processing. min(t+1) is a minimum data in real-time running original data at the second time, and max(t+1) is a maximum data in real-time running original data at the second time.

[0128] Next, a sum of a product of the first coefficient and the first minimum data and a product of the second coefficient and the second minimum data is determined as an updated normalized interval minimum data value, that is, x min (t+1) =a1*xmin t +a2*min(t+1), x min (t+1) is the updated normalized interval minimum data value, a1 is a first coefficient, and a2 is a second coefficient. The values of the first coefficient and the second coefficient can be set according to actual needs, but need to satisfy a1>a2 and a1+a2=1. In an example, a1 can be set as 0.9 and a2 can be set as 0.1.

[0129] On the other hand, it can also be determined that the sum of the product of the first coefficient and the first maximum data and the product of the second coefficient and the second maximum data is the updated normalized interval maximum data value, that is, x max (t+1) =a1*x max t +a2*max(t+1), x max (t +1) is the updated normalized interval maximum data value.

[0130] Then, the updated normalized interval can be determined based on the updated normalized interval minimum data value and the updated normalized interval maximum data value, and the sample data can be mapped to the updated normalized interval to update the device operation data. In an example, the sample data can be mapped to the updated normalized interval using a min-max normalization processing method to update the device operation data.

[0131] In some embodiments of the present application, updating the fault prediction model based on the updated device operation data can be retraining the fault prediction model based at least on the updated device operation data.

[0132] Figure 8 is a flowchart of a method for updating a fault prediction model based on updated device operation data provided by an embodiment of the present application. As Figure 8 shown, the method includes the following steps:

[0133] In step S801, the distribution offset degree of the updated device operation data is determined.

[0134] In step S802, in response to determining that the distribution offset degree is greater than the preset offset degree threshold, the learning rate of the fault prediction model is increased.

[0135] In step S803, the fault prediction model is retrained based on the updated device operation data and the increased learning rate to obtain an updated fault prediction model.

[0136] In some embodiments of the present application, the distribution offset degree of the updated equipment operation data can be determined first. In an example, the distribution offset degree can be the KL (Kullback-Leibler) divergence. If it is determined that the KL divergence is greater than a preset offset degree threshold, the learning rate of the fault prediction model is increased, and the fault prediction model is retrained based on the updated equipment operation data and the updated learning rate to obtain an updated fault prediction model.

[0137] For example, if it is determined that the KL divergence of the updated equipment operation data exceeds 0.1, the learning rate of the fault prediction model can be increased from 0.001 to 0.01. The learning rate can be a maximum threshold of the model learning rate.

[0138] Figure 9 is a flowchart of another device fault prediction method provided by an embodiment of the present application. As shown in Figure 9 When performing device fault prediction, data can be collected first by a data collection module, and the collected data can be preprocessed. A fault prediction model can be constructed based on the preprocessed data, the fault prediction model at least including an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer, and an output layer, and the model parameters can be trained based on the preprocessed data to obtain a trained fault prediction model. Next, the trained fault prediction model can be used to perform fault prediction based on real-time equipment operation data, and the prediction results can be evaluated and fed back. If it is determined based on the feedback results that incremental learning of the fault prediction model is needed, the collected data is updated using real-time equipment operation data, and then the updated data is preprocessed, and the model is updated using an incremental learning method, thereby obtaining an updated fault prediction model. Finally, the updated fault prediction model can be used to predict device faults.

[0139] By adopting the technical solutions provided by the embodiments of the present application, the fusion of the spatio-temporal features and scene parameters of industrial equipment is realized, the fault prediction model can be quickly updated to adapt the model to new scenes, the accuracy and real-time performance of fault prediction are improved, and the model performance is optimized through continuous feedback, thereby reducing maintenance costs and improving equipment safety.

[0140] All the optional technical solutions described above can be combined to form optional embodiments of the present application, which will not be described here.

[0141] The following is an apparatus embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the apparatus embodiments of the present application, please refer to the method embodiments of the present application.

[0142] Figure 10 is a schematic diagram of a device fault prediction apparatus provided by an embodiment of the present application. As shown in Figure 10As shown, the apparatus comprises:

[0143] The acquisition module 1001 is configured to acquire device operation data, the device operation data comprising at least one of real-time operation data of the device at a first time and historical operation data of the device.

[0144] The construction module 1002 is configured to construct a fault prediction model based on the device operation data, the fault prediction model comprising at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer, and an output layer; wherein the input layer is used to input the device operation data; the convolutional neural network layer is used to extract features from the device operation data to obtain initial feature data; the gated recurrent unit and the attention layer are used to capture long-term dependencies and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is used to obtain a fault prediction result based on the target feature data.

[0145] The prediction module 1003 is configured to acquire real-time operation data of the device at a second time, input the real-time operation data at the second time into the fault prediction model, and obtain a fault prediction result; wherein the second time is later than the first time.

[0146] The update module 1004 is configured to acquire review label information of the fault prediction result, and in response to determining that a target index of the fault prediction model is less than a preset index threshold based on the fault prediction result and the review label information, update the device operation data based on the real-time operation data at the second time, and update the fault prediction model based on the updated device operation data; wherein the target index comprises at least one of a fault prediction accuracy and a fault prediction score.

[0147] The prediction module 1003 is further configured to predict a fault of the device using the updated fault prediction model.

[0148] According to the technical scheme provided in the embodiments of the present application, by acquiring device operation data, constructing a fault prediction model based on the device operation data, predicting real-time operation data of the device at a second time using the constructed fault prediction model to obtain a fault prediction result, and in response to determining that a target index of the fault prediction model is less than a preset index threshold based on the fault prediction result and review label information of the fault prediction result, updating the device operation data based on the real-time operation data at the second time, updating the fault prediction model based on the updated device operation data, and finally predicting a fault of the device using the updated fault prediction model, real-time data of the device can be analyzed and processed, and an incremental learning method is used to construct a fault prediction model, model parameters are updated in real time to improve prediction accuracy and real-time performance, maintenance costs are reduced, and device safety is improved.

[0149] In some embodiments, the convolution kernel size and the number of convolution layers of the convolutional neural network layer are determined as follows: obtaining a data type of the device operation data; the data type includes high-frequency sensor data and low-frequency sensor data, the high-frequency sensor data is data with a sampling frequency greater than or equal to a preset frequency threshold, and the low-frequency sensor data is data with a sampling frequency less than the preset frequency threshold; in response to determining that the data type is high-frequency sensor data, setting the convolution kernel size of the convolutional neural network layer to a first size and the number of convolution layers to a first number; in response to determining that the data type is low-frequency sensor data, setting the convolution kernel size of the convolutional neural network layer to a second size and the number of convolution layers to a second number; wherein the first size is less than the second size, and the first number is greater than the second number.

[0150] In some embodiments, the number of hidden units and the number of attention mechanism heads of the gated recurrent unit are determined as follows: determining a working condition fluctuation coefficient of the device based on the standard deviation of the device operation data; in response to determining that the working condition fluctuation coefficient is greater than or equal to a preset fluctuation threshold, setting the number of hidden units of the gated recurrent unit to a first value and the number of attention mechanism heads to a second value; in response to determining that the working condition fluctuation coefficient is less than the preset fluctuation threshold, setting the number of hidden units of the gated recurrent unit to a third value and the number of attention mechanism heads to a fourth value; wherein the first value is greater than the third value, and the second value is greater than the fourth value.

[0151] In some embodiments, the attention layer determines the relevance weight of each time step in the initial feature data to the current prediction time based on a regularization factor; the regularization factor is determined as follows: obtaining the number of sensors and the number of fault types corresponding to the device operation data; multiplying the ten logarithm value of the sum of the number of sensors and the number of fault types by a preset weight coefficient to obtain the regularization factor.

[0152] In some embodiments, the fault prediction accuracy is determined as follows: based on the fault prediction result and the review label information, determining the number of correctly predicted positive categories, the number of correctly predicted negative categories, the number of incorrectly predicted positive categories, and the number of incorrectly predicted negative categories; determining the sum of the number of correctly predicted positive categories and the number of correctly predicted negative categories as a first sum; determining the sum of the number of correctly predicted positive categories, the number of correctly predicted negative categories, the number of incorrectly predicted positive categories, and the number of incorrectly predicted negative categories as a second sum; determining the quotient of the first sum and the second sum as the fault prediction accuracy.

[0153] In some embodiments, the fault prediction score is determined as follows: based on the fault prediction result and the review label information, determining the prediction precision and the prediction recall; determining twice the product of the prediction precision and the prediction recall as a first product; determining the sum of the prediction precision and the prediction recall as a third sum; determining the quotient of the first product and the third sum as the fault prediction score.

[0154] In some embodiments, the device operation data is pre-processed data, and the pre-processing at least includes min-max normalization processing; updating the device operation data based on the real-time operation data at the second time point includes: respectively obtaining a first minimum data in the device operation data and a second minimum data in the real-time operation data at the second time point, and a first maximum data in the device operation data and a second maximum data in the real-time operation data at the second time point; wherein the first minimum data and the first maximum data are normalized data; determining a sum of a product of a first coefficient and the first minimum data and a product of a second coefficient and the second minimum data as an updated normalized interval minimum data value; determining a sum of a product of the first coefficient and the first maximum data and a product of the second coefficient and the second maximum data as an updated normalized interval maximum data value; determining an updated normalized interval based on the updated normalized interval minimum data value and the updated normalized interval maximum data value, and mapping sample data to the updated normalized interval to update the device operation data; wherein the sample data includes the real-time operation data at the second time point, or includes the device operation data and the real-time operation data at the second time point; and the first coefficient is greater than the second coefficient.

[0155] In some embodiments, updating the fault prediction model based on the updated device operation data includes: determining a distribution offset degree of the updated device operation data; in response to determining that the distribution offset degree is greater than a preset offset threshold, increasing a learning rate of the fault prediction model; and retraining the fault prediction model based on the updated device operation data and the increased learning rate to obtain an updated fault prediction model.

[0156] It should be understood that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0157] Figure 11 is a schematic diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 11 the electronic device 11 of this embodiment includes a processor 1101, a memory 1102, and a computer program 1103 stored in the memory 1102 and executable on the processor 1101. The processor 1101 implements the steps in each of the above method embodiments when executing the computer program 1103. Alternatively, the processor 1101 implements the functions of each module / unit in each of the above device embodiments when executing the computer program 1103.

[0158] The electronic device 11 can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The electronic device 11 can include but is not limited to the processor 1101 and the memory 1102. Those skilled in the art can understand that Figure 11The electronic device 11 is merely an example and does not limit the electronic device 11, and can include more or less components than illustrated, or different components.

[0159] The processor 1101 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc.

[0160] The memory 1102 can be an internal storage unit of the electronic device 11, for example, a hard disk or a memory of the electronic device 11. The memory 1102 can also be an external storage device of the electronic device 11, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 11. The memory 1102 can also include both the internal storage unit and the external storage device of the electronic device 11. The memory 1102 is used to store computer programs and other programs and data required by the electronic device.

[0161] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or in the form of software functional unit.

[0162] The integrated modules / units, if implemented in the form of software functional units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can be executed by a processor to implement the steps of the above-mentioned various method embodiments. The computer program can include computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc.

[0163] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A device failure prediction method characterized by, The method comprises: obtaining equipment operation data, the equipment operation data comprising at least one of real-time operation data of the equipment at a first time and historical operation data of the equipment; constructing a fault prediction model based on the equipment operation data, the fault prediction model comprising at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer, and an output layer; wherein the input layer is configured to input the equipment operation data; the convolutional neural network layer is configured to extract features from the equipment operation data to obtain initial feature data; the gated recurrent unit and the attention layer are configured to capture long-term dependencies and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is configured to obtain a fault prediction result based on the target feature data; obtaining real-time operation data of the equipment at a second time, inputting the real-time operation data of the equipment at the second time into the fault prediction model, and obtaining a fault prediction result; wherein the second time is later than the first time; obtaining review label information of the fault prediction result, and in response to determining that a target index of the fault prediction model is less than a preset index threshold based on the fault prediction result and the review label information, updating the equipment operation data based on the real-time operation data of the equipment at the second time, and updating the fault prediction model based on the updated equipment operation data; wherein the target index comprises at least one of a fault prediction accuracy and a fault prediction score; predicting a fault of the equipment using the updated fault prediction model.

2. The method of claim 1, wherein, The size of the convolution kernel and the number of convolution layers of the convolutional neural network layer are determined in the following manner: obtaining a data type of the equipment operation data; the data type comprises high-frequency sensor data and low-frequency sensor data, the high-frequency sensor data being data with a sampling frequency greater than or equal to a preset frequency threshold, and the low-frequency sensor data being data with a sampling frequency less than the preset frequency threshold; in response to determining that the data type is high-frequency sensor data, setting the size of the convolution kernel of the convolutional neural network layer to a first size and the number of convolution layers to a first number; in response to determining that the data type is low-frequency sensor data, setting the size of the convolution kernel of the convolutional neural network layer to a second size and the number of convolution layers to a second number; wherein the first size is less than the second size, and the first number is greater than the second number.

3. The method of claim 1, wherein, The number of hidden units and the number of attention mechanism heads of the gated recurrent unit are determined in the following manner: determining a working condition fluctuation coefficient of the equipment based on a standard deviation of the equipment operation data; in response to determining that the working condition fluctuation coefficient is greater than or equal to a preset fluctuation threshold, setting the number of hidden units of the gated recurrent unit to a first value and the number of attention mechanism heads to a second value; in response to determining that the working condition fluctuation coefficient is less than the preset fluctuation threshold, setting the number of hidden units of the gated recurrent unit to a third value and the number of attention mechanism heads to a fourth value; wherein the first value is greater than the third value, and the second value is greater than the fourth value.

4. The method of claim 1, wherein, The attention layer determines a relevance weight of each time step in the initial feature data and a current prediction time based on a regularization factor; and the regularization factor is determined in the following manner: Obtain the number of sensors and the number of fault types corresponding to the equipment operation data; Multiply the ten logarithm value of the sum of the number of sensors and the number of fault types by a preset weight coefficient to obtain the regularization factor.

5. The method of claim 1, wherein, The fault prediction accuracy is determined in the following manner: Determine the number of correctly predicted positive classes, the number of correctly predicted negative classes, the number of incorrectly predicted positive classes, and the number of incorrectly predicted negative classes based on the fault prediction result and the review label information; Determine the sum of the number of correctly predicted positive classes and the number of correctly predicted negative classes as a first sum value; Determine the sum of the number of correctly predicted positive classes, the number of correctly predicted negative classes, the number of incorrectly predicted positive classes, and the number of incorrectly predicted negative classes as a second sum value; Determine the quotient of the first sum value and the second sum value as the fault prediction accuracy.

6. The method of claim 1, wherein, The fault prediction score is determined in the following manner: Determine the prediction precision and the prediction recall rate based on the fault prediction result and the review label information; Determine the product of twice the prediction precision and the prediction recall rate as a first product; Determine the sum of the prediction precision and the prediction recall rate as a third sum value; Determine the quotient of the first product and the third sum value as the fault prediction score.

7. The method of claim 1, wherein, The equipment operation data is preprocessed data, and the preprocessing at least includes minimum-maximum normalization processing; The updating of the equipment operation data based on the real-time operation data at the second time comprises: Obtain a first minimum data in the equipment operation data and a second minimum data in the real-time operation data at the second time, and a first maximum data in the equipment operation data and a second maximum data in the real-time operation data at the second time; wherein the first minimum data and the first maximum data are normalized data; Determine the sum of the product of a first coefficient and the first minimum data and the product of a second coefficient and the second minimum data as an updated normalized interval minimum data value; Determine the sum of the product of the first coefficient and the first maximum data and the product of the second coefficient and the second maximum data as an updated normalized interval maximum data value; Determine an updated normalized interval based on the updated normalized interval minimum data value and the updated normalized interval maximum data value, map sample data to the updated normalized interval, and update the equipment operation data; Wherein, the sample data includes the real-time operation data at the second time, or includes the equipment operation data and the real-time operation data at the second time; and the first coefficient is greater than the second coefficient.

8. The method of claim 1, wherein, The updating of the fault prediction model based on the updated equipment operation data comprises: Determine a distribution offset degree of the updated equipment operation data; In response to determining that the distribution offset degree is greater than a preset offset threshold, increase the learning rate of the fault prediction model; and In response to determining that the distribution offset degree is less than or equal to the preset offset threshold, maintain the learning rate of the fault prediction model. retrain the fault prediction model based on the updated equipment operation data and the increased learning rate to obtain an updated fault prediction model.

9. An apparatus failure prediction device characterized by comprising: The method comprises the steps of: an acquisition module configured to acquire equipment operation data, the equipment operation data comprising at least one of real-time operation data of equipment at a first time and historical operation data of the equipment; a construction module configured to construct a fault prediction model based on the equipment operation data, the fault prediction model comprising at least an input layer, a convolutional neural network layer, a gated recurrent unit, an attention layer, and an output layer; wherein the input layer is configured to input the equipment operation data; the convolutional neural network layer is configured to perform feature extraction on the equipment operation data to obtain initial feature data; the gated recurrent unit and the attention layer are configured to capture long-term dependency relationships and adjust correlation weights of the initial feature data to obtain target feature data; and the output layer is configured to obtain a fault prediction result based on the target feature data; a prediction module configured to acquire real-time operation data of the equipment at a second time, input the real-time operation data of the equipment at the second time into the fault prediction model, and obtain a fault prediction result; wherein the second time is later than the first time; an updating module configured to acquire review label information of the fault prediction result, and in response to determining that a target index of the fault prediction model is less than a preset index threshold based on the fault prediction result and the review label information, update the equipment operation data based on the real-time operation data of the equipment at the second time, and update the fault prediction model based on the updated equipment operation data; wherein the target index comprises at least one of a fault prediction accuracy and a fault prediction score; the prediction module is further configured to predict a fault of the equipment using the updated fault prediction model.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 8. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 8.

Images