Intelligent data fusion method and system for fault prediction

Abstract

The invention relates to the technical field of data fusion, and discloses an intelligent data fusion method and system for fault prediction, and the method comprises the steps: obtaining historical operation data and real-time operation data of target equipment, and carrying out the data fusion of the historical operation data and the real-time operation data, and obtaining first fusion data; inputting the first fusion data into a convolutional neural network to obtain first feature data, and inputting the first fusion data into a long short-term memory network to obtain second feature data; and finally, fusing the first feature data and the second feature data to obtain second fused data which is used for fault prediction of the target equipment. According to the method, the fusion data of the historical operation data and the real-time operation data are respectively input into the convolutional neural network and the long-short-term memory network to obtain the feature data, and the feature data are fused for fault prediction, so that the prediction model can process the spatial features and the time sequence features at the same time, and the obtained fault prediction result is more accurate; and a guarantee is provided for safe operation of equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and in particular to an intelligent data fusion method and system for fault prediction. Background Art

[0002] Existing fault prediction systems typically rely solely on historical or real-time data. Historical data provides context for fault occurrences, but due to its limited timeliness, it may not reflect the current device status. Real-time data, while providing immediate status, lacks sufficient historical context, resulting in incomplete feature representation. Existing methods rely on either historical or real-time data, failing to leverage information from both. This incomplete feature representation leads to high false positive and false negative rates in fault prediction. Traditional models also lack sufficient flexibility to address new fault modes, resulting in poor adaptability. Summary of the Invention

[0003] In view of this, the present invention provides an intelligent data fusion method and system for fault prediction, which can effectively solve the problem of incomplete feature representation caused by a single data source through a data fusion strategy, so as to improve the accuracy of fault model prediction.

[0004] In a first aspect, the present invention provides an intelligent data fusion method for fault prediction, the method comprising:

[0005] Acquire historical operation data and real-time operation data of the target device, and fuse the historical operation data and the real-time operation data to obtain first fused data;

[0006] Inputting the first fused data into a convolutional neural network to obtain first feature data, where the first feature data is used to characterize the spatiotemporal features of the target device operation data;

[0007] Inputting the first fused data into a long short-term memory network to obtain second feature data, where the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period;

[0008] The first feature data and the second feature data are fused to obtain second fused data for use in fault prediction of a target device.

[0009] The intelligent data fusion method for fault prediction provided by the embodiment of the present invention inputs the fused data of historical and real-time operation data into the convolutional neural network and long short-term memory network models respectively, obtains the spatiotemporal characteristics characterizing the operation data of the target equipment and the change characteristics of the trend or pattern within a preset time period, and fuses these two types of feature data for model prediction. Through a comprehensive data fusion strategy, the prediction model can simultaneously process spatial features and time series features, and the obtained fault prediction results are more accurate, providing a guarantee for the safe operation of the equipment.

[0010] In an optional embodiment, the historical operation data and the real-time operation data both include corresponding sensor data at timestamps of the same sampling period, and fusing the historical operation data and the real-time operation data to obtain first fused data includes:

[0011] Set the weights of historical operation data and real-time operation data;

[0012] The historical operation data and the real-time operation data corresponding to the timestamps of the same sampling period are fused based on a weighted fusion method to obtain first fused data.

[0013] The reliability and relevance of historical and real-time data may vary for different devices, or for the same device at different operating stages. This embodiment of the present invention uses a weighted fusion approach to flexibly adjust weights based on actual conditions, thereby adapting to various complex device operating environments and data characteristics. This allows the resulting first fused data to better reflect the device's actual operating status, providing a higher-quality data foundation for subsequent deep learning analysis, thereby improving the accuracy and reliability of fault prediction.

[0014] In an optional implementation, inputting the first fused data into a convolutional neural network to obtain first feature data includes:

[0015] Construct a convolutional neural network architecture and initialize network parameters, including setting the number of convolution kernels according to the sensor data type and the time step according to the sensor data sampling period;

[0016] A convolutional neural network is constructed based on historical operation data training, and the fusion data of different sensor data types are input into the trained convolutional neural network to obtain the feature map data corresponding to each sensor data type;

[0017] Based on the feature map data corresponding to all sensor data types, the feature matrix is obtained as the first feature data F CNN＝ (N, M, 1), where N represents the number of sensor data types, M represents the time step, K represents the convolution kernel size, L represents the number of channels for each feature, and 1 represents the number of channels for each feature.

[0018] The embodiment of the present invention integrates the feature graph data corresponding to different sensor data types into a feature matrix as the first feature data, thereby realizing a multi-dimensional comprehensive feature representation of the equipment's operating status. This integration method can comprehensively reflect the operating conditions of different aspects of the equipment, avoiding the limitations of single sensor data. Each sensor data type provides information about the equipment's operation from a specific perspective. By combining their feature graph data, a more complete description of the equipment's operating status is formed. In subsequent fault prediction or equipment status analysis, such comprehensive features can improve the reliability of analysis and judgment, reduce the risk of misjudgment and missed judgment, and provide a strong decision-making basis for equipment maintenance and management.

[0019] In an optional embodiment, inputting the first fused data into a long short-term memory network to obtain second feature data includes:

[0020] Convert the fused data of different sensor data types into a three-dimensional tensor;

[0021] Construct a long short-term memory network and initialize the network parameters, and then train the network based on historical data to obtain a trained long short-term memory network model;

[0022] The fusion data of different sensor data types are input into the set long short-term memory network model respectively, and the second feature data is expressed as F LSTM =(batch_size,t,hidden_dim), where batch_size represents the number of samples of input data, t represents the time step length, and hidden_dim represents the dimension of the hidden state.

[0023] This embodiment of the present invention integrates fused data from different sensor types into a trained LSTM network model, enabling comprehensive analysis of multi-sensor data. Each sensor data reflects the device's operating status from a different perspective. By processing this data through the LSTM network, it is possible to integrate information from different sensors, discovering their interrelationships and coordinated change patterns, thereby providing a more comprehensive and accurate description of the device's operating status and further improving the reliability and accuracy of fault prediction.

[0024] In an optional implementation, fusing the first feature data and the second feature data to obtain second fused data includes:

[0025] Aligning the time steps of the first feature data and the second feature data;

[0026] Convert the first feature data into data with the same format as the second feature data to obtain FCNN1＝ (batch_size,t,N);

[0027] The first feature data after format conversion is fused with the second feature data to obtain second fused data F = (batch_size, t, N + hidden_dim).

[0028] The embodiment of the present invention aligns the time steps of the first feature data and the second feature data, ensuring that the time scales reflected by the two data are consistent during the fusion process. The first feature data is converted to the same format as the second feature data, eliminating fusion barriers that may be caused by data format differences. The resulting second fused data integrates two different types of features, fully utilizing the spatiotemporal information in the first feature data and the trend change information in the second feature data. This fusion of multi-dimensional information can more comprehensively describe the operating status of the equipment. Compared with single feature data, it provides a richer basis for fault prediction, thereby improving the reliability and accuracy of fault prediction and better meeting the needs of equipment maintenance and management.

[0029] In an optional embodiment, the method further includes: setting the corresponding operating status labels of the second fusion data to form a training set for training a preset deep learning model, and using the trained model to predict equipment failures.

[0030] The second fused data used in the embodiment of the present invention integrates multiple information such as the spatiotemporal characteristics from the convolutional neural network and the trend change characteristics of the long short-term memory network. Using such rich fused data for training can enable the deep learning model to fully understand the various conditions and change patterns of equipment operation. The trained model can better capture the subtle signs and complex change trends before equipment failure occurs. Combining the second fused data with the corresponding operating status label to form a training set can enable the preset deep learning model to directly learn for the equipment failure prediction task, thereby making more accurate judgments in fault prediction and reducing false alarms and missed alarms.

[0031] In a second aspect, the present invention provides an intelligent data fusion system for fault prediction, comprising:

[0032] A first fusion module is used to obtain historical operation data and real-time operation data of the target device, and fuse the historical operation data and the real-time operation data to obtain first fused data;

[0033] a first feature data acquisition module, configured to input the first fused data into a convolutional neural network to obtain first feature data, wherein the first feature data is used to characterize the spatiotemporal features of the target device operation data;

[0034] a second feature data acquisition module, configured to input the first fused data into a long short-term memory network to obtain second feature data, wherein the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period;

[0035] The second fusion module is used to fuse the first feature data and the second feature data to obtain second fused data for fault prediction of the target device.

[0036] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to thereby execute the intelligent data fusion method for fault prediction of the above-mentioned first aspect or any corresponding embodiment thereof.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the intelligent data fusion method for fault prediction of the above-mentioned first aspect or any corresponding embodiment thereof.

[0038] In a fifth aspect, the present invention provides a computer program product comprising computer instructions for causing a computer to execute the intelligent data fusion method for fault prediction according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 is a flow chart of an intelligent data fusion method for fault prediction according to an embodiment of the present invention;

[0041] Figure 2 is a flow chart of another intelligent data fusion method for fault prediction according to an embodiment of the present invention;

[0042] Figure 3 is a structural block diagram of an intelligent data fusion system for fault prediction according to an embodiment of the present invention;

[0043] Figure 4 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0045] Existing fault models often rely on a single data source during training, which fails to fully capture the complexity of equipment operating conditions, thus limiting the effectiveness and accuracy of fault prediction. Based on this, the present invention provides a data fusion strategy based on deep learning that can significantly improve the performance of intelligent fault prediction systems.

[0046] Specifically, an embodiment of the present invention provides an embodiment of an intelligent data fusion method for fault prediction. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0047] In this embodiment, an intelligent data fusion method for fault prediction is provided. Figure 1 As shown, the process includes the following steps:

[0048] S101 , acquiring historical operation data and real-time operation data of a target device, and fusing the historical operation data and the real-time operation data to obtain first fused data.

[0049] This embodiment of the present invention captures readings from various sensors used to monitor industrial equipment as operational data, including temperature, pressure, vibration, noise, and current. Historical operational data represents the equipment's operational data from the past day, sampled 60 times per hour, resulting in 24*60 = 1440 records. An example of a captured data record includes the following: timestamp: 2024-01-01 12:00:00; collector data: temperature: 80°C, pressure: 2.0 MPa, vibration: 0.05 mm / s, noise: 55 dB, and current: 5.0 A. If data anomalies are detected, the fault type, such as abnormal current, is also recorded.

[0050] For real-time operation data, the current operation indicator data of industrial equipment is obtained from the data acquisition end using the same data recording method as the historical operation data, that is, the above-mentioned 5 indicator data and the timestamp of each set of data. The obtained real-time data is formed into a real-time data stream, and 60 data items are obtained within one hour.

[0051] The historical operation data and the real-time operation data are further integrated. The embodiment of the present invention adopts the weighted average method to perform data integration. According to the reliability and timeliness of the data, the historical operation data and the real-time operation data corresponding to the timestamp of the same sampling period are combined to assign a weight W to the historical operation data. h and real-time operation data F real-time Assign weight W r , the weighted historical operating data F history and real-time operation data F real-time The corresponding features are added to obtain the first fused data F fused =W h *F history +W r *F real-time .

[0052] Regarding the selection of weights, for example, real-time operation data is given a higher weight such as 0.7 because it reflects the current actual situation; for historical operation data, attenuation weighting is performed according to the time distance, and the closer the historical data is to the current time, the higher the weight, and the overall weight is 0.3.

[0053] The reliability and relevance of historical and real-time data may vary for different devices, or for the same device at different operating stages. This embodiment of the present invention uses a weighted fusion approach to flexibly adjust weights based on actual conditions, thereby adapting to various complex device operating environments and data characteristics. This allows the resulting first fused data to better reflect the device's actual operating status, providing a higher-quality data foundation for subsequent deep learning analysis, thereby improving the accuracy and reliability of fault prediction.

[0054] S102: Input the first fused data into a convolutional neural network to obtain first feature data, where the first feature data is used to characterize the spatiotemporal features of the target device operation data.

[0055] Specifically, the embodiment of the present invention uses a convolutional neural network model trained with historical operating data to identify information related to spatiotemporal features extracted from the first fused data. This eliminates the need for manually designing complex feature extraction rules, greatly reducing labor costs and reliance on domain expert experience, and improving the efficiency and accuracy of feature extraction. For example, for complex equipment operation data, which may contain measurement values of multiple sensors at different time and spatial locations, the convolutional neural network can automatically discover hidden correlation patterns and dynamic change patterns, such as the spatiotemporal distribution characteristics of parameters such as temperature and pressure of the equipment at different operating stages. For example, in equipment monitoring on industrial automation production lines, the convolutional neural network can extract the characteristics of equipment operating status changes in different production batches and different time periods, as well as the mutual influence relationship between different components of the equipment, providing strong support for equipment fault prediction and performance optimization.

[0056] S103: Input the first fused data into a long short-term memory network to obtain second feature data, where the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period.

[0057] The embodiment of the present invention trains a long short-term memory network using historical operating data. Since the trained long short-term memory network can use its unique gating mechanism to selectively remember or forget historical information, it can accurately identify long-term dependencies in the data. For complex dynamic changes that may occur during equipment operation, such as the gradual decline of equipment performance and abnormal fluctuations before sudden failures, the LSTM network can respond in a timely manner and extract corresponding features. By extracting accurate trend and pattern change features, it provides strong support for subsequent tasks such as equipment failure prediction and performance optimization.

[0058] S104: Fusing the first feature data and the second feature data to obtain second fused data for use in fault prediction of the target device.

[0059] In the embodiment of the present invention, the first feature data characterizes the spatiotemporal features of the equipment operation data, and the second feature data reflects the changing features of the trend or pattern of the equipment operation data within a preset time period. The fused data obtained by fusing them integrates these two key information, which can more comprehensively describe the operation status of the equipment, make up for the limitations of single feature data in fault prediction, and help to more accurately identify potential faults of the equipment. This is because the occurrence of faults is often closely related to the spatiotemporal operation mode of the equipment and long-term trend changes; the fused second fused data enables the fault prediction model to adapt to more complex equipment operation environments and working condition changes. Whether it is a small fluctuation during normal operation of the equipment or a performance change under different working conditions, it can be more accurately analyzed and predicted through comprehensive feature information, thereby improving the generalization ability of the model in practical applications, so that it can be applied to equipment fault prediction of different types and different working conditions, and has a broader application prospect.

[0060] The embodiment of the present invention also provides another intelligent data fusion method for fault prediction, such as Figure 2 As shown, the following steps are included:

[0061] S201 , acquiring historical operation data and real-time operation data of a target device, and fusing the historical operation data and the real-time operation data to obtain first fused data.

[0062] For details, see step S101 and no further details will be given here.

[0063] S202: Input the first fused data into a convolutional neural network to obtain first feature data, where the first feature data is used to characterize the spatiotemporal features of the target device operation data.

[0064] The embodiment of the present invention performs step S202, which specifically includes the following steps:

[0065] S2021, build the convolutional neural network architecture and initialize the network parameters, including: setting the number of convolution kernels of the convolutional neural network according to the sensor data type and setting the time step according to the sensor data sampling period.

[0066] Specifically, a convolutional neural network is constructed, which includes multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer uses an appropriate convolution kernel size (such as 3x3, etc.) and step size to perform a convolution operation on the first fused data. The parameters of the convolutional neural network (such as convolution kernel weights, biases, etc.) are initialized. The embodiment of the present invention is based on five types of data: temperature, pressure, vibration, noise, and current. For each data type, the number of convolution kernels is set according to the complexity and diversity of its data characteristics. For example, for temperature data, if it changes relatively smoothly during the operation of the equipment, fewer convolution kernels (such as 16-32) may be set to extract its basic features; for vibration data, since it may contain rich frequency components and complex patterns, more convolution kernels (such as 64-128) may be required to fully capture its features. Then, a large number of labeled target equipment operation data samples (the labeled information may be whether the equipment is operating normally, the type of fault, etc.) are used for training.

[0067] The network parameters are continuously adjusted through the back-propagation algorithm so that the network can accurately extract information related to the spatiotemporal features from the first fused data. During the training process, appropriate loss functions (such as the cross entropy loss function for classification tasks and the mean square error loss function for regression tasks) and optimizers (such as stochastic gradient descent, Adam optimizer, etc.) are selected to optimize network performance. The convolutional neural network trained with a large amount of data has strong adaptability and generalization capabilities. It can process different types of target device operating data as long as the basic features and patterns of the data are within the training range. Even if the device operating environment changes or some new situations arise, the network can perform reasonable feature extraction and judgment based on the learned knowledge. For example, for devices of different models but with similar working principles, a convolutional neural network that has been properly adjusted and trained can quickly adapt to and extract effective spatiotemporal features, thereby achieving effective monitoring and analysis of the device's operating status.

[0068] S2022, training a convolutional neural network based on historical operation data, and inputting the fusion data of different sensor data types into the trained convolutional neural network to obtain feature map data corresponding to each sensor data type;

[0069] Specifically, the number of samples of each sensor data type is 1440, so the shape of the input data is (5, 1440, 1), that is: the number of time steps is 1440, the number of features is 5, the number of channels of each feature is 1, and the time series of each feature has only one value, that is, a one-dimensional sequence.

[0070] For each data input sequence X, the calculation formula of the output feature map is:

[0071]

[0072] in:

[0073] X: Input historical data vector.

[0074] W: weight of the convolution kernel.

[0075] b: Bias term.

[0076] F: Output feature map.

[0077] i: Indicates the current position in the input sequence.

[0078] j: represents the index of the current convolution kernel.

[0079] In one embodiment, if the temperature sensor data sequence X = [85, 80, 75, ...] is input, the convolution kernel W = [1, 0, -1] and the bias term b = 0, the first eigenvalue is calculated:

[0080] F 1，j =W 1，j 85+W 2，j 80+W 3，j 75+b j

[0081] That is: F[1,1]=(1*85)+(0*80)+(-1*75)+0=85-75=10. Similar operations are performed on each feature (temperature, pressure, vibration, etc.) to obtain the convolution output of each feature.

[0082] S2023, based on the feature map data corresponding to all sensor data types, obtain a feature matrix as the first feature data F CNN =(N-K+1,M-K+1,L), where N is the number of sensor data types, M is the time step, K is the convolution kernel size, and L is the number of channels for each feature.

[0083] Specifically, since the number of sensor data types in the embodiment of the present invention is 5, the acquisition time step M is 1440, and the convolution kernel size K is 3, the convolution results of all features are merged into a feature map with a shape of (3, 1438, 1).

[0084] S203: Input the first fused data into a long short-term memory network to obtain second feature data, where the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period.

[0085] Specifically, executing step S203 specifically includes the following steps:

[0086] S2031, converting the fused data of different sensor data types into a three-dimensional tensor;

[0087] Specifically, like the convolutional neural network, the sensor data types are 5 and the historical duration is 1440. The historical data of each data collection is formatted as follows:

[0088] X sensor =[x t-1440 ,x t-1439 ,...,x t-1 ]

[0089] Among them, X sensor It is a tensor containing the historical data from each collector. The data is merged into a three-dimensional tensor with a shape of (batch_size, 1440, 5), where batch_size refers to the basic dimension or initial size of the input data.

[0090] S2032, constructing a long short-term memory network and initializing network parameters, and training based on historical data to obtain a trained long short-term memory network model.

[0091] Specifically, a long short-term memory (LSTM) network is first constructed. Its core structure includes an input gate, a forget gate, an output gate, and a memory unit. The number of network layers and the number of hidden units in each layer are then determined, generally based on the complexity of the data and the task requirements. For example, for complex device operation data, a multi-layer LSTM may be required to better capture long-term dependencies in the data. The input layer receives the first fused data, whose dimensions must match the number of data features. Parameters such as the weights and biases of the LSTM network are initialized using random initialization or methods based on specific distributions (such as Xavier initialization or He initialization). Training is then performed using a large number of historical operation data samples from the target device. Each sample contains a time series of the first fused data and the corresponding device operation status label (such as normal, fault type, etc.). During training, the backpropagation through time (BPTT) algorithm is used to calculate gradients and update network parameters. An appropriate loss function (such as the mean squared error loss function) and optimizer (such as the Adam optimizer) are used to minimize the error between the predicted results and the true label.

[0092] S2033: Input the fusion data of different sensor data types into the set long short-term memory network model respectively, and obtain the second feature data represented by F LSTM =(batch_size,t,hidden_dim), where batch_size represents the number of samples of input data, t represents the time step length, and hidden_dim represents the dimension of the hidden state.

[0093] Specifically, the dimension of the hidden state is a hyperparameter, which is generally set to 64 or 128, which is only used as an example and is not limited to this.

[0094] S204: Fusing the first feature data and the second feature data to obtain second fused data for fault prediction of the target device.

[0095] Specifically, executing step S204 includes the following steps:

[0096] S2041: Align the time steps of the first feature data and the second feature data.

[0097] Specifically, the embodiment of the present invention uses a connection operation to fuse the first feature data and the second feature data. Before performing feature fusion, it is necessary to ensure that the number of time steps output by the CNN and LSTM is consistent. Since the CNN output is (3, 1438, 1) and the LSTM output is (batch_size, 1440, hidden_dim), the time steps need to be aligned. Usually, the CNN output is padded or truncated to make it consistent with the number of time steps output by the LSTM. For example, the last 1438 time steps of the LSTM output can be intercepted, or the CNN time step can be extended to 1440 using zero padding. The embodiment of the present invention uses the latter, i.e., 1440.

[0098] S2042, converting the first characteristic data into data in the same format as the second characteristic data to obtain F CNN1＝ (batch_size,t,N).

[0099] Specifically, the first dimension of the fused feature data is determined by the output of LSTM, which represents the number of samples in each batch, i.e., batch_size. In addition, multiple convolution kernels are introduced when CNN and LSTM are fused. The number of convolution kernels is related to the number of features. In this embodiment of the present invention, 5 kernels are selected, so the third dimension of the CNN output becomes 5.

[0100] S2043 , fusing the first feature data after format conversion with the second feature data to obtain second fused data F=(batch_size, t, N+hidden_dim).

[0101] Specifically, the output features of CNN and LSTM are spliced in the feature dimension. Since the feature F extracted by CNN CNN =(batch_size,1440,5), LSTM extracted features F LSTM＝ (batch_size,1440,hidden_dim), so F=concat(F CNN , F LSTM ), and the final fused features are obtained with a shape of (batch_size, 1440, 5+hidden_dim).

[0102] The fused feature F will contain rich information from five different indicator dimensions, which can more comprehensively reflect the operating status and fault characteristics of the equipment. The feature output of each dimension indicator is and After fusion, it can be expressed as:

[0103]

[0104] For equipment failure prediction, many failure modes may manifest as sudden changes, periodic fluctuations, or chronic trend changes, which all involve the combination of spatiotemporal features and time-dependent features. final Used in subsequent fault prediction models, the model can extract more accurate signals from multi-dimensional and multi-level information, thereby effectively improving the accuracy of fault detection and prediction.

[0105] Furthermore, the method provided by the embodiment of the present invention also includes:

[0106] S205, the second fusion data is set with the corresponding operation status label to form a training set for training the preset deep learning model, and the trained model is used to predict the failure of the equipment. final Set a label (fault or normal), for example, label 0 indicates normal status, label 1 indicates fault. For example:

[0107] F final =[F temp =75,F pressure =2.1,F vibration =0.02,F noise =45,F current =4.9] label Y=0

[0108] Furthermore, we use the multi-layer perceptron (MLP) model as an example to predict equipment failure, and transform the feature tensor F final The training dataset is composed of the label Y, and the model is trained using the training dataset. The validation dataset is used to adjust the model’s hyperparameters (such as the number of neurons in the hidden layer, the learning rate, etc.). The trained model is deployed to the actual environment, and new real-time running data is continuously passed into the model to generate the feature tensor F. final , which is then fed into a fault prediction model. The model outputs a prediction (whether the device is faulty). If the prediction is faulty (labeled as 1), an alert is sent, initiating a warning mechanism to alert maintenance personnel to perform inspections or maintenance. Using a deep learning model (MLP), historical data can be effectively combined with real-time data to analyze the time series characteristics of device operating status and predict potential faults. This significantly improves the device's fault identification capabilities, providing early warnings and avoiding equipment downtime.

[0109] This embodiment also provides an intelligent data fusion system for fault prediction. This system is used to implement the above-mentioned embodiments and preferred implementations, and details already described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0110] This embodiment provides an intelligent data fusion system for fault prediction, such as Figure 3 Shown, including:

[0111] The first fusion module 301 is used to obtain historical operation data and real-time operation data of the target device, and fuse the historical operation data and the real-time operation data to obtain first fused data;

[0112] A first feature data acquisition module 302 is configured to input the first fused data into a convolutional neural network to obtain first feature data, where the first feature data is used to characterize the spatiotemporal features of the target device operation data;

[0113] A second feature data acquisition module 303 is configured to input the first fused data into a long short-term memory network to obtain second feature data, where the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period;

[0114] The second fusion module 304 is configured to fuse the first feature data and the second feature data to obtain second fused data for use in fault prediction of a target device.

[0115] In some optional implementations, the first fusion module 301 includes:

[0116] A weight setting unit is used to set the weights of historical operation data and real-time operation data;

[0117] The first fusion unit is used to fuse the historical operation data and the real-time operation data corresponding to the timestamps of the same sampling period based on a weighted fusion method to obtain first fused data.

[0118] In some optional implementations, the first feature data acquisition module 302 includes:

[0119] Convolutional neural network initialization unit, used to build the convolutional neural network architecture and initialize the network parameters, including: setting the number of convolution kernels of the convolutional neural network according to the sensor data type and setting the time step according to the sensor data sampling period;

[0120] A feature graph data acquisition unit is used to train a convolutional neural network based on historical operation data, and input the fusion data of different sensor data types into the trained convolutional neural network to obtain feature graph data corresponding to each sensor data type;

[0121] The first feature data acquisition unit is used to obtain the feature matrix as the first feature data F based on the feature map data corresponding to all sensor data types. CNN＝ (N, M, 1), where N represents the number of sensor data types, M represents the time step, K represents the convolution kernel size, L represents the number of channels for each feature, and 1 represents the number of channels for each feature.

[0122] In some optional implementations, the second feature data acquisition module 303 includes:

[0123] A data format conversion unit is used to convert the fusion data of different sensor data types into a three-dimensional tensor;

[0124] The long short-term memory network training unit is used to construct a long short-term memory network and initialize network parameters, and obtain a trained long short-term memory network model based on historical data training;

[0125] The second feature data acquisition unit is used to input the fusion data of different sensor data types into the set long short-term memory network model respectively, and obtain the second feature data represented by F LSTM =(batch_size,t,hidden_dim), where batch_size represents the number of samples of input data, t represents the time step length, and hidden_dim represents the dimension of the hidden state.

[0126] In some optional implementations, the second fusion module 304 includes:

[0127] A time alignment unit, configured to align the time steps of the first feature data and the second feature data;

[0128] The characteristic data format conversion unit is used to convert the first characteristic data into data with the same format as the second characteristic data to obtain F CNN1＝ (batch_size,t,N);

[0129] The second data fusion unit is used to fuse the first feature data after format conversion with the second feature data to obtain second fused data F=(batch_size, t, N+hidden_dim).

[0130] In some optional embodiments, the system further comprises:

[0131] The prediction application unit is used to set the operating status labels corresponding to the second fusion data to form a training set for training a preset deep learning model, and use the trained model to predict equipment failures.

[0132] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0133] The intelligent data fusion system for fault prediction in this embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0134] The embodiment of the present invention also provides a computer device having the above Figure 3 The intelligent data fusion system for fault prediction is shown.

[0135] See also Figure 4 , Figure 4 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 4 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 4 A processor 10 is taken as an example.

[0136] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0137] The memory 20 stores instructions that can be executed by at least one processor 10, so as to enable at least one processor 10 to execute the method shown in the above embodiment.

[0138] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0139] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0140] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0141] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor central control system or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0142] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0143] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. An intelligent data fusion method for fault prediction, characterized in that: include: Acquire historical operation data and real-time operation data of the target device, and fuse the historical operation data and the real-time operation data to obtain first fused data; Inputting the first fused data into a convolutional neural network to obtain first feature data, where the first feature data is used to characterize the spatiotemporal features of the target device operation data; Inputting the first fused data into a long short-term memory network to obtain second feature data, where the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period; The first feature data and the second feature data are fused to obtain second fused data for use in fault prediction of a target device.

2. The method according to claim 1, characterized in that The historical operation data and the real-time operation data both include corresponding sensor data at timestamps of the same sampling period, and the data fusion of the historical operation data and the real-time operation data to obtain first fused data includes: Set the weights of historical operation data and real-time operation data; The historical operation data and the real-time operation data corresponding to the timestamps of the same sampling period are fused based on a weighted fusion method to obtain first fused data.

3. The method according to claim 1, characterized in that Inputting the first fused data into a convolutional neural network to obtain first feature data includes: Construct a convolutional neural network architecture and initialize network parameters, including setting the number of convolution kernels according to the sensor data type and the time step according to the sensor data sampling period; A convolutional neural network is constructed based on historical operation data training, and the fusion data of different sensor data types are input into the trained convolutional neural network to obtain the feature map data corresponding to each sensor data type; Based on the feature map data corresponding to all sensor data types, the feature matrix is obtained as the first feature data F CNN＝ (N ， M ， 1), where N represents the number of sensor data types, M represents the time step, K represents the convolution kernel size, and L represents the number of channels for each feature.

4. The method according to claim 2, characterized in that Inputting the first fused data into a long short-term memory network to obtain second feature data includes: Convert the fused data of different sensor data types into a three-dimensional tensor; Construct a long short-term memory network and initialize the network parameters, and then train the network based on historical data to obtain a trained long short-term memory network model; The fusion data of different sensor data types are input into the set long short-term memory network model respectively, and the second feature data is expressed as F LSTM =(batch_size,t,hidden_dim), where batch_size represents the number of samples of input data, t represents the time step length, and hidden_dim represents the dimension of the hidden state.

5. The method according to claim 4, characterized in that The fusing the first feature data and the second feature data to obtain second fused data includes: Aligning the time steps of the first feature data and the second feature data; Convert the first feature data into data with the same format as the second feature data to obtain F CNN1＝ (batch_size,t,N); The first feature data after format conversion is fused with the second feature data to obtain second fused data F = (batch_size, t, N + hidden_dim).

6. The method according to claim 1, wherein Also includes: The operating status labels corresponding to the second fusion data are set to form a training set for training a preset deep learning model, and the trained model is used to predict equipment failures.

7. An intelligent data fusion system for fault prediction, characterized in that: include: A first fusion module is used to obtain historical operation data and real-time operation data of the target device, and fuse the historical operation data and the real-time operation data to obtain first fused data; a first feature data acquisition module, configured to input the first fused data into a convolutional neural network to obtain first feature data, wherein the first feature data is used to characterize the spatiotemporal features of the target device operation data; a second feature data acquisition module, configured to input the first fused data into a long short-term memory network to obtain second feature data, wherein the second feature data is used to characterize the changing characteristics of the trend or pattern of the target device operation data within a preset time period; The second fusion module is used to fuse the first feature data and the second feature data to obtain second fused data for fault prediction of the target device.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the intelligent data fusion method for fault prediction according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the intelligent data fusion method for fault prediction according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the intelligent data fusion method for fault prediction according to any one of claims 1 to 6.

Images