Feed pump fault diagnosis and model training method, equipment, medium and program product

Abstract

The embodiment of the invention provides a feed pump fault diagnosis and model training method and device, a medium and a program product, and the model training method comprises the steps: obtaining a fault diagnosis model obtained by training a sample set of a source feed pump, the model comprising a feature extractor and a fault classifier; acquiring a sample set of a target water feeding pump; inputting the sample set of the target feed pump into a feature extractor to obtain a feature vector of each sample of the target feed pump, and inputting the feature vector into a fault classifier to obtain a training output tag of each sample of the target feed pump; respectively calculating the feature vector of each sample of the source feed pump and the target feed pump and the outer product of the training output label to obtain a first joint feature and a second joint feature; inputting the first joint feature and the second joint feature into a domain discriminator to obtain domain discrimination results of all samples; determining an adversarial loss based on the domain discrimination results of all the samples and a preset domain label; and adjusting network parameters of the feature extractor based on the adversarial loss.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of artificial intelligence, and in particular to a water supply pump fault diagnosis and model training method, device, medium and program product. BACKGROUND

[0002] As an important component of auxiliary machines of coal-fired power plants, the water supply pump directly affects the safety and operation efficiency of the power plant, so it is of great significance to diagnose the fault of the water supply pump. At present, the fault diagnosis model is usually trained to diagnose the fault of the water supply pump, and the training of the fault diagnosis model usually depends on a large number of labeled sample data. However, in actual application, the water supply pump to be diagnosed often only has limited sample data.

[0003] The common practice of the prior art is to first train a fault diagnosis model on a water supply pump with sufficient samples, and then directly apply the trained fault diagnosis model to the water supply pump to be diagnosed. However, different water supply pumps differ in pump type, operating conditions, load, flow rate and temperature gradient, etc., which makes it difficult to directly apply the fault diagnosis model trained in the water supply pump with sufficient samples to the water supply pump to be diagnosed.

[0004] Therefore, there is an urgent need for a method that can quickly build a fault diagnosis model for the water supply pump to be diagnosed using the trained fault diagnosis model under the condition that the sample of the water supply pump to be diagnosed is limited, so as to improve the accuracy and stability of the fault detection of the water supply pump to be diagnosed. SUMMARY

[0005] Embodiments of the present application provide a water supply pump fault diagnosis and model training method, device, medium and program product, which can quickly build a fault diagnosis model for a target pump based on a fault diagnosis model of a source water supply pump, significantly reduce sample dependence, and improve the accuracy and stability of the fault diagnosis model.

[0006] In a first aspect, the water supply pump fault diagnosis model training method provided by embodiments of the present application comprises:

[0007] obtaining a fault diagnosis model trained using a sample set of a source water supply pump, the fault diagnosis model comprising a feature extractor and a fault classifier;

[0008] obtaining a sample set of a target water supply pump, the number of samples in the sample set of the target water supply pump being less than the number of samples in the sample set of the source water supply pump;

[0009] inputting the sample set of the target water supply pump into the feature extractor for feature extraction to obtain a feature vector of each sample of the target water supply pump, inputting the feature vector of each sample of the target water supply pump into the fault classifier for fault classification to obtain a training output label corresponding to each sample of the target water supply pump;

[0010] An outer product of the feature vector of each sample of the source water supply pump and the training output label is calculated to obtain a first joint feature, and an outer product of the feature vector of each sample of the target water supply pump and the training output label is calculated to obtain a second joint feature;

[0011] The first joint feature and the second joint feature are input into the domain discriminator to perform domain discrimination to obtain domain discrimination results of all samples;

[0012] An adversarial loss is determined based on the domain discrimination results of all samples and preset domain labels of all samples;

[0013] The network parameters of the feature extractor are adjusted based on the adversarial loss.

[0014] In a second aspect, the water supply pump fault diagnosis method provided by the embodiments of the present application comprises:

[0015] The current data of the multi-dimensional features of the target water supply pump are obtained, and the current data of the multi-dimensional features of the target water supply pump are input into the fault diagnosis model trained by any embodiment of the present application;

[0016] The feature extractor is used for feature extraction to obtain the feature vector of the target water supply pump;

[0017] The fault classifier is used for fault classification to obtain the predicted fault label of the target water supply pump.

[0018] In a third aspect, the water supply pump fault diagnosis model training device provided by the embodiments of the present application comprises:

[0019] A model acquisition module is configured to acquire a fault diagnosis model trained by using a sample set of a source water supply pump, wherein the fault diagnosis model comprises a feature extractor and a fault classifier;

[0020] A sample acquisition module is configured to acquire a sample set of a target water supply pump, wherein the number of samples in the sample set of the target water supply pump is less than the number of samples in the sample set of the source water supply pump;

[0021] A sample training module is configured to input the sample set of the target water supply pump into the feature extractor for feature extraction to obtain a feature vector of each sample of the target water supply pump, and input the feature vector of each sample of the target water supply pump into the fault classifier for fault classification to obtain a training output label corresponding to each sample of the target water supply pump;

[0022] A joint feature calculation module is configured to calculate an outer product of the feature vector of each sample of the source water supply pump and the training output label to obtain a first joint feature, and calculate an outer product of the feature vector of each sample of the target water supply pump and the training output label to obtain a second joint feature;

[0023] The domain discrimination module is configured to input the first joint feature and the second joint feature into a domain discriminator to perform domain discrimination, and obtain domain discrimination results of all samples.

[0024] The adversarial loss determination module is configured to determine an adversarial loss based on the domain discrimination results of all samples and preset domain labels of all samples.

[0025] The network parameter adjustment module is configured to adjust network parameters of the feature extractor based on the adversarial loss.

[0026] In a fourth aspect, the water pump fault diagnosis device provided by the embodiments of the present application comprises:

[0027] The data acquisition module is configured to acquire current multi-dimensional feature data of a target water pump, and input the current multi-dimensional feature data of the target water pump into a fault diagnosis model trained according to any of the embodiments of the present application.

[0028] The feature vector extraction module is configured to perform feature extraction by using the feature extractor to obtain a feature vector of the target water pump.

[0029] The fault classification module is configured to perform fault classification by using the fault classifier to obtain a predicted fault label of the target water pump.

[0030] In a fifth aspect, the electronic device provided by the embodiments of the present application comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the water pump fault diagnosis model training method according to any of the embodiments of the present application or the water pump fault diagnosis method according to any of the embodiments of the present application when executing the program.

[0031] In a sixth aspect, the computer readable storage medium provided by the embodiments of the present application stores a computer program, and the program is executed by a processor to implement the water pump fault diagnosis model training method according to any of the embodiments of the present application or the water pump fault diagnosis method according to any of the embodiments of the present application.

[0032] In a seventh aspect, the computer program product provided by the embodiments of the present application comprises a computer program, and the computer program is executed by a processor to implement the water pump fault diagnosis model training method according to any of the embodiments of the present application or the water pump fault diagnosis method according to any of the embodiments of the present application.

[0033] In the embodiment of the present application, the fault diagnosis model trained by using the sample set of the source water supply pump is obtained, the fault diagnosis model includes a feature extractor and a fault classifier, which can provide a basic model for subsequent target domain fault diagnosis; the sample set of the target water supply pump is obtained, the number of samples in the sample set of the target water supply pump is less than the number of samples in the sample set of the source water supply pump, which can obtain a limited sample set of the target water supply pump and provide a data basis for transfer learning; the sample set of the target water supply pump is input into the feature extractor for feature extraction, the feature vector of each sample of the target water supply pump is obtained, the feature vector of each sample of the target water supply pump is input into the fault classifier for fault classification, and the training output label of each sample of the target water supply pump is obtained correspondingly, which can extract effective features from the limited samples of the target water supply pump and generate a preliminary prediction result, thereby providing a basis for subsequent domain adaptation and model optimization; the cross product of the feature vector and the training output label of each sample of the source water supply pump is calculated to obtain a first joint feature, and the cross product of the feature vector and the training output label of each sample of the target water supply pump is calculated to obtain a second joint feature, which can jointly represent the feature vector and the prediction information, enhance the distinguishability of the sample in the domain discriminator, and provide effective input for adversarial training; the first joint feature and the second joint feature are input into the domain discriminator for domain discrimination to obtain the domain discrimination result of all samples, which can provide a basis for determining the adversarial loss; the adversarial loss is determined based on the domain discrimination result of all samples and the preset domain label of all samples, which can provide an optimization direction for the feature extractor and enhance the generalization ability of the model in the target domain; the network parameters of the feature extractor are adjusted based on the adversarial loss, which can make the feature vector extracted by the feature extractor difficult to distinguish the sample source, weaken the feature difference between the source domain and the target domain, and thus improve the accuracy and stability of the fault diagnosis of the target water supply pump. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0035] Figure 1 is a flowchart of the water pump fault diagnosis model training method provided by the embodiment of the present application;

[0036] Figure 2 is another flowchart of the water pump fault diagnosis model training method provided by the embodiment of the present application;

[0037] Figure 3 is a flowchart of the water pump fault diagnosis method provided by the embodiment of the present application;

[0038] Figure 4 is a structural schematic diagram of a water supply pump fault diagnosis model training device provided by an embodiment of the present application;

[0039] Figure 5 is a structural schematic diagram of a water supply pump fault diagnosis device provided by an embodiment of the present application;

[0040] Figure 6 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts should fall within the scope of the present application.

[0042] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to only those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0043] Figure 1 is a flowchart of a water supply pump fault diagnosis model training method provided by an embodiment of the present application. The water supply pump fault diagnosis model training method provided by the embodiment of the present application can be applied to the scenario of constructing a target water supply pump fault diagnosis model by using a source water supply pump fault diagnosis model in the case of limited target water supply pump samples. The water supply pump fault diagnosis model training method can be executed by the water supply pump fault diagnosis model training device provided by the embodiment of the present application, which can be realized in the form of software and / or hardware. In a specific embodiment, the device can be integrated in an electronic device, which can be a computer, a server, or the like. The following embodiments will be described by taking the example of integrating the water supply pump fault diagnosis model training device in an electronic device. Referring to Figure 1 , the water supply pump fault diagnosis model training method of the present embodiment can include the following steps:

[0044] Step 101, obtain a fault diagnosis model trained by using a sample set of a source water supply pump, the fault diagnosis model including a feature extractor and a fault classifier.

[0045] The sample set of the source water supply pump refers to a collection of a plurality of sample data collected from the source water supply pump, each sample including multi-dimensional feature historical data and a corresponding preset fault label; wherein the source water supply pump refers to a water supply pump with a large number of sample data and corresponding preset fault labels. The fault diagnosis model refers to a fault diagnosis model trained based on the sample set of the source water supply pump, a model for diagnosing water pump faults, which includes two parts: a feature extractor and a fault classifier. The feature extractor refers to a network structure for automatically extracting deep feature representation from the multi-dimensional features of the input sample, which can usually be constructed using a convolutional neural network. The fault classifier refers to a classification network structure for judging the fault category based on the feature vector output by the feature extractor, which can usually be constructed using a deep neural network.

[0046] Specifically, a large number of sample data of the source water supply pump are obtained, and the model is trained according to the sample data of the source water supply pump to obtain a fault diagnosis model capable of fault recognition, which consists of two parts: one part is a feature extractor for extracting features from input data, and the other part is a fault classifier for judging fault categories according to features. That is, a fault diagnosis model combined by a feature extractor and a fault classifier is pre-trained by a sample set of a source water supply pump to provide initial model parameters for subsequent migration to a target water supply pump.

[0047] Step 102, obtain a sample set of a target water supply pump, the number of samples in the sample set of the target water supply pump being less than the number of samples in the sample set of the source water supply pump.

[0048] The sample set of the target water supply pump refers to a collection of a plurality of sample data collected from the target water supply pump, each sample including multi-dimensional feature historical data; wherein the multi-dimensional feature historical data usually includes high-frequency signal features and low-frequency signal features, and the target water supply pump refers to a water supply pump with only a small number of sample data and most of the sample data lacking preset fault labels. The number of samples refers to the number of samples included in the sample set.

[0049] Specifically, sample data of the target water supply pump are obtained to obtain a sample set of the target water supply pump, the number of samples in the set being significantly less than the number of samples obtained from the source water supply pump, i.e. the sample data available for training of the target water supply pump is significantly insufficient.

[0050] Step 103: Input the sample set of the target water pump into the feature extractor for feature extraction to obtain the feature vector of each sample of the target water pump. Input the feature vector of each sample of the target water pump into the fault classifier for fault classification to obtain the training output label of each sample of the target water pump.

[0051] A feature vector is a vector generated by a feature extractor after extracting features from a sample. Training output labels are the probabilities of each fault category generated by a fault classifier after classifying the feature vectors of the input samples.

[0052] Specifically, the sample data of the target water pump is input into the feature extractor, which processes each sample data to obtain a feature vector that can represent the sample data. These feature vectors are then input into the fault classifier, which predicts the fault category of each sample, thereby obtaining the training output label corresponding to each sample of the target water pump.

[0053] Step 104: Calculate the outer product of the feature vector of each sample of the source water pump and the training output label to obtain the first joint feature; calculate the outer product of the feature vector of each sample of the target water pump and the training output label to obtain the second joint feature.

[0054] The outer product refers to the result of combining two vectors according to the rules of the outer product operation. The first joint feature is the result of performing the outer product operation between the sample feature vector of the source water pump and its training output label. The second joint feature is the result of performing the outer product operation between the sample feature vector of the target water pump and its training output label.

[0055] For example, the feature vector and training output label of the m-th sample of the source water pump are respectively represented as f s (m) and g s (m), where f represents the feature vector and training output label of the m-th sample of the target water pump, respectively. t (m) and g t (m), where m represents the index of a sample in the sample set, used to identify the m-th sample. The outer product rule is represented as ⊗, then the first joint feature of the m-th sample is t. s =f s (m)⊗g s (m), the second joint feature is t t =f t (m)⊗g t (m).

[0056] Step 105: Input the first joint feature and the second joint feature into the domain discriminator for domain discrimination to obtain the domain discrimination results of all samples.

[0057] A domain discriminator is a discriminative network that distinguishes between input joint features and target domain samples. A binary classification multilayer perceptron is typically used as the domain discriminator. The domain discrimination result is the output probability value obtained by the domain discriminator based on the input joint features, indicating whether the sample is more likely to belong to the source or target domain.

[0058] Specifically, the first joint feature from the source domain and the second joint feature from the target domain are both input into the domain discriminator. The domain discriminator determines whether each sample is a source domain sample or a target domain sample, and finally outputs a domain discrimination result. If 0 represents that the joint feature belongs to the source domain sample and 1 represents that the joint feature belongs to the target domain sample, then the closer the domain discrimination result is to 1, the more likely the sample belongs to the target domain.

[0059] Continuing with the previous example, let's assume the domain discriminator is... Then the first joint feature t s The second joint feature is t t The input domain discriminator performs domain discrimination, and can obtain the domain discrimination results for all samples. , where t=t s or t=t t The expression corresponding to the domain discrimination is: , , ,in, and Both represent activation functions. and This represents two layers of feature vectors. , and Represents the weight matrix. , and Both indicate bias.

[0060] Step 106: Determine the adversarial loss based on the domain discrimination results of all samples and the preset domain labels of all samples.

[0061] Preset domain labels refer to the true domain category labels pre-set for each sample during the training phase, used to indicate whether the sample actually belongs to the source domain or the target domain. Adversarial loss refers to the loss value formed by the difference between the domain discrimination result and the preset domain labels. It is used to measure the accuracy of the domain discriminator in classifying the sample domain and is used for adversarial training to encourage the feature extractor to learn domain-independent features.

[0062] Specifically, adversarial loss is calculated based on the domain discrimination result given by the domain discriminator for each sample and the preset domain label of the sample. This loss is used to train the domain discriminator, enabling it to accurately distinguish between source domain samples and target domain samples. At the same time, a gradient reversal mechanism is used to apply to the feature extractor, making the features extracted by the feature extractor more domain invariant, thereby improving the model's generalization ability in the target domain.

[0063] Continuing with the previous example, assuming d represents the preset domain label, then based on the domain discrimination result output by the domain discriminator for each sample... Given the preset domain label d for each sample, the adversarial loss is determined as follows: ,in, Indicates the number of samples in the source domain. Indicates the number of samples in the target domain. Represents the source domain sample set, Let x represent the target domain sample set, and let x represent any sample from the source domain sample set and the target domain sample set.

[0064] Step 107: Adjust the network parameters of the feature extractor based on adversarial loss.

[0065] Network parameters refer to the weight and bias parameters of each layer in the feature extractor that need to be updated through training. They determine how the feature extractor extracts features from the sample data.

[0066] Specifically, the calculated adversarial loss is used to propagate the adversarial gradient generated by the domain discriminator back to the feature extractor by multiplying it by a negative coefficient through the gradient inversion layer. The Adam optimizer is then used to perform backpropagation updates on the network parameters of the feature extractor, enabling the feature extractor to gradually learn feature representations with domain invariance. This reduces the feature differences between the source and target domains and improves the model's generalization ability in the target domain.

[0067] In this embodiment, a fault diagnosis model trained using a sample set of source feedwater pumps is obtained. This model includes a feature extractor and a fault classifier, providing a foundational model for subsequent target domain fault diagnosis. A sample set of target feedwater pumps is also obtained; the number of samples in the target feedwater pump sample set is less than the number of samples in the source feedwater pump sample set, thus providing a limited sample set for transfer learning. The sample set of target feedwater pumps is input into the feature extractor for feature extraction, yielding a feature vector for each sample. This feature vector is then input into the fault classifier for fault classification, resulting in a training output label for each sample. This process extracts effective features from the limited sample set of target feedwater pumps and generates preliminary prediction results, providing a basis for subsequent domain adaptation and model optimization. Finally, the features of each sample from the source feedwater pump are calculated. The first joint feature is obtained by multiplying the feature vector and the training output label. The second joint feature is obtained by multiplying the feature vector and the training output label for each sample of the target water pump. This allows for the joint representation of feature vectors and prediction information, enhancing the discriminability of samples in the domain discriminator and providing effective input for adversarial training. The first and second joint features are then input into the domain discriminator for domain discrimination, yielding the domain discrimination results for all samples. This provides a basis for determining the adversarial loss. The adversarial loss is determined based on the domain discrimination results of all samples and the preset domain labels of all samples, providing optimization direction for the feature extractor and enhancing the model's generalization ability in the target domain. Adjusting the network parameters of the feature extractor based on the adversarial loss makes it difficult for the feature vectors extracted by the feature extractor to distinguish the sample source, reducing the feature differences between the source and target domains, thereby improving the accuracy and stability of fault diagnosis for the target water pump.

[0068] Figure 2 This is another flowchart illustrating the water pump fault diagnosis model training method provided in this embodiment of the invention, as shown below. Figure 2 As shown, the water pump fault diagnosis model training method in this embodiment may include:

[0069] Step 201: Obtain the fault diagnosis model trained using the sample set of the source water pump. The fault diagnosis model includes a feature extractor and a fault classifier.

[0070] The sample set of the source water pump includes multiple samples of the source water pump. Each sample of the source water pump includes multi-dimensional feature historical data of the source water pump and preset fault labels.

[0071] The multidimensional historical feature data of the source feedwater pump refers to a collection of multi-source historical signal data used to characterize the state of the source feedwater pump. This includes statistical features of high-frequency signals (e.g., root mean square value, peak value, kurtosis, and waveform factor of vibration signals) and mean features of low-frequency signals (e.g., mean values ​​of motor stator current, temperature, pressure, and flow rate). The preset fault labels for the source feedwater pump are predefined labels that identify the fault types corresponding to the samples of the source feedwater pump, used to supervise the training of the fault classifier.

[0072] Specifically, the sample set of the source water pump consists of multiple samples of the source water pump. Each sample contains multi-dimensional feature historical data recording the status of the source water pump and preset fault labels used to label the actual fault type of the sample.

[0073] Step 202: Obtain the sample set of the target water pump, where the number of samples in the sample set of the target water pump is less than the number of samples in the sample set of the source water pump.

[0074] The sample set of the target feedwater pump includes multiple samples of the target feedwater pump. Each sample of the target feedwater pump includes multi-dimensional feature historical data of the target feedwater pump, or includes multi-dimensional feature historical data of the target feedwater pump and preset fault labels.

[0075] The multidimensional feature historical data of the target feedwater pump refers to a set of multi-source historical signal data used to characterize the state of the target feedwater pump, including the statistical characteristics of high-frequency signals (such as the root mean square value, peak value, kurtosis, and waveform factor of vibration signals) and the mean characteristics of low-frequency signals (such as the mean values ​​of motor stator current signals, temperature signals, pressure signals, and flow signals). The preset fault labels of the target feedwater pump refer to predefined labels that identify the fault types corresponding to the samples of the target feedwater pump, used to supervise the training of the fault classifier.

[0076] Specifically, the sample set of the target feedwater pump consists of multiple samples of the target feedwater pump. Each sample contains at least multi-dimensional feature historical data recording the status of the target feedwater pump, and a small number of samples contain preset fault labels for labeling the true fault type of the sample.

[0077] Optionally, the historical data for multidimensional features can be obtained through the following methods:

[0078] Using a preset sliding window with a preset step size, the signal is obtained by sliding the signal from the historical signal data to obtain multiple historical window signals. The historical signal data includes historical high-frequency signals and historical low-frequency signals.

[0079] Calculate the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal, and then concatenate the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal to obtain multiple multidimensional feature historical data.

[0080] A preset sliding window refers to a pre-defined fixed-length time window (e.g., 20 seconds) used to sequentially slide through historical signal data according to a preset step size, capturing historical signal data within the window's coverage area each time as the historical window signal. The preset step size refers to the pre-defined time distance the sliding window moves each time, for example, 5 seconds. Historical signal data refers to multi-source time-series signal data collected from the water pump, used to characterize the pump's state. If the water pump is the source pump, this historical signal data is used to generate multi-dimensional feature historical data for the source pump; if the water pump is the target pump, this historical signal data is used to generate multi-dimensional feature historical data for the target pump. The historical window signal refers to the historical signal data covered by the sliding window. Historical high-frequency signals refer to high-frequency signals collected within the sliding window, such as vibration signals, used to calculate statistical characteristics such as energy, peak value, kurtosis, and waveform factor. Historical low-frequency signals refer to low-frequency signals collected within the sliding window, such as current, temperature, pressure, and flow rate, used to calculate mean characteristics. Statistical characteristics refer to numerical indicators extracted from high-frequency signals, such as root mean square value, peak value, kurtosis, and waveform factor, used to characterize the amplitude and morphological properties of the signal. Mean characteristics refer to the average value calculated from low-frequency signals, used to reflect the overall level of the low-frequency signal.

[0081] Specifically, a preset sliding window is used to slide and extract signals from historical signal data at a preset step size, resulting in multiple historical window signals. The historical signal data includes high-frequency and low-frequency signals. The statistical characteristics of the high-frequency signal and the mean characteristics of the low-frequency signal in each historical window signal are calculated, and they are concatenated to obtain multiple multidimensional feature historical data. Each historical window signal corresponds to one multidimensional feature historical data.

[0082] For example, the historical signal data is set to include vibration signals. Motor stator current signal Temperature signal Pressure signal Flow signal ,in, Generally, it is a high-frequency signal (frequency is...). ), other signals It is a low-frequency signal (frequency is First, determine the length of the sliding window. and step length (like , ), then the first The signal span covered by each window is Based on this, we can obtain There are several windows, each with a set of characteristics. For the high-frequency signal in the m-th historical window... Calculate the following statistical characteristics: root mean square value Peak , cliff and waveform factor ,in, For the low-frequency signal in the m-th historical window Calculate the mean characteristics of all low-frequency signals to obtain The statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in the m-th historical window are concatenated to obtain the m-th multidimensional feature historical data. In addition, after obtaining the first Then the signal can be Normalization to [-1, 1], i.e. ,in, For signal The minimum value, For signal The maximum value, These are the normalized values. By calculating the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal using this method and then concatenating them, multiple multidimensional historical feature data can be obtained. If there are multiple vibration measurement points (such as motor end / pump end, X / Y direction), the features can be calculated using the same format and then directly concatenated.

[0083] In addition, before performing sliding signal retrieval from historical signal data using a preset sliding window, any signal in the historical signal data can also be processed. Outliers are removed or replaced to reduce the impact of data bias on the training of the fault diagnosis model. Specifically, this is achieved through the formula... Calculate signal At any moment Statistic Through formula Calculate variance ,in, The average value of the signal. The length of the signal sequence. The significance level is selected. (e.g., 0.05), used to control the confidence level for outlier testing. The critical value of the statistic is calculated using the following formula. ,in, The t-distribution is based on degrees of freedom T⁻² and significance level. The critical value below, when At that time, the corresponding observation value Removed as outlier or used Replacement.

[0084] Step 203: Input the sample set of the target water pump into the feature extractor for feature extraction to obtain the feature vector of each sample of the target water pump. Input the feature vector of each sample of the target water pump into the fault classifier for fault classification to obtain the training output label of each sample of the target water pump.

[0085] Step 204: Calculate the outer product of the feature vector of each sample of the source water pump and the training output label to obtain the first joint feature; calculate the outer product of the feature vector of each sample of the target water pump and the training output label to obtain the second joint feature.

[0086] Step 205: Input the first joint feature and the second joint feature into the domain discriminator for domain discrimination to obtain the domain discrimination results of all samples.

[0087] Step 206: Determine the adversarial loss based on the domain discrimination results of all samples and the preset domain labels of all samples.

[0088] Step 207: Determine the source domain loss based on the training output labels of all samples from the source water pump and the preset fault labels of all samples.

[0089] For example, the training output label of the m-th sample from the source water pump is represented as... The default fault label is represented as The source domain loss is ,in, Indicates the number of samples in the source domain. Indicates the overall fault category. Indicates the first in the source domain The window parameter corresponds to the first The value represents the type of fault; if the fault belongs to that type, the value is 1; otherwise, the value is 0. Indicates the first in the source domain The window parameter corresponds to the first Class failure probability.

[0090] Step 208: Determine the target domain loss based on the preset fault labels of all samples of the target water pump.

[0091] For example, the training output label of the m-th sample of the target water pump is represented as... The target domain loss is ,in, Indicates the number of samples in the target domain. Indicates the first in the target domain The window parameter corresponds to the first Class failure probability; this value is available when the target label exists. replace.

[0092] Step 209: Determine the overall loss based on the source domain loss, target domain loss, and adversarial loss.

[0093] Continuing with the previous example, the overall loss is ,in Indicates source domain loss. Indicates the target domain loss. Indicating resistance to loss, Weighting coefficients, used to counteract loss, are employed to adjust the importance of domain adaptation learning in the overall optimization. The weighting coefficient for the target domain loss is used to adjust the proportion of the target domain distribution constraint in the overall optimization.

[0094] Step 210: Adjust the network parameters of the feature extractor based on the overall loss.

[0095] Specifically, the overall gradient is determined by calculating the overall loss and then passed to the feature extractor. The network parameters of the feature extractor are then updated based on the overall gradient.

[0096] Optionally, the network parameters of the feature extractor are adjusted based on the overall loss, including: determining the overall gradient passed to the feature extractor based on the overall loss, the overall gradient including the source domain gradient, the target domain gradient and the adversarial gradient; multiplying the adversarial gradient by a preset negative coefficient to obtain the inverted adversarial gradient; and adjusting the network parameters of the feature extractor based on the source domain gradient, the target domain gradient and the inverted adversarial gradient.

[0097] Source domain gradient refers to the gradient generated by backpropagation of the source domain loss. Target domain gradient refers to the gradient generated by backpropagation of the target domain loss. Adversarial gradient refers to the gradient obtained by backpropagation of the adversarial loss, used to prompt the feature extractor to generate domain-indistinguishable features. Inverted adversarial gradient refers to multiplying the adversarial gradient by a preset negative coefficient (e.g., ...). The gradient obtained afterward is used to update the feature extractor in the direction of the deception domain discriminator.

[0098] Specifically, the feature extractor is updated by the combined action of the source domain gradient, the target domain gradient, and the reversed adversarial gradient, making it difficult for the feature vector output by the feature extractor to distinguish whether it comes from the source water pump or the target water pump, while also making it easier for the fault classifier to distinguish different fault types.

[0099] In this embodiment, a fault diagnosis model trained using a sample set of source water pumps is obtained. The fault diagnosis model includes a feature extractor and a fault classifier. The sample set of source water pumps includes multiple samples from the source water pumps. Each sample from the source water pumps includes multi-dimensional feature historical data and a preset fault label, providing a basic model for subsequent target domain fault diagnosis. A sample set of target water pumps is also obtained, where the number of samples in the target water pump sample set is less than the number of samples in the source water pump sample set. The target water pump sample set includes multiple samples from the target water pumps. Each sample includes historical multidimensional features of the target feedwater pump, or historical multidimensional features of the target feedwater pump and a preset fault label, enabling the acquisition of a limited sample set of target feedwater pumps, providing a data foundation for transfer learning. The sample set of target feedwater pumps is input into a feature extractor for feature extraction, obtaining a feature vector for each sample of the target feedwater pump. The feature vector of each sample of the target feedwater pump is then input into a fault classifier for fault classification, corresponding to the training output label for each sample of the target feedwater pump. This allows for the extraction of effective features from a limited sample of the target feedwater pump and the generation of preliminary prediction results, providing a basis for subsequent domain... This provides a basis for adaptation and model optimization; the first joint feature is obtained by calculating the outer product of the feature vector and the training output label for each sample of the source water pump, and the second joint feature is obtained by calculating the outer product of the feature vector and the training output label for each sample of the target water pump. This allows for the joint representation of feature vectors and prediction information, enhancing the discriminability of samples in the domain discriminator and providing effective input for adversarial training; the first and second joint features are input into the domain discriminator for domain discrimination, obtaining the domain discrimination results for all samples, which provides a basis for subsequently determining the adversarial loss; based on the domain discrimination results of all samples and the preset domain labels of all samples... The adversarial loss is determined by labeling, which can provide optimization direction for the feature extractor and enhance the model's generalization ability in the target domain. The source domain loss is determined based on the training output labels of all samples from the source water pump and the preset fault labels of all samples. The target domain loss is determined based on the preset fault labels of all samples from the target water pump. The overall loss is determined based on the source domain loss, target domain loss, and adversarial loss. The network parameters of the feature extractor are adjusted based on the overall loss, which enables the model to consider the source domain gradient, target domain gradient, and adversarial gradient simultaneously during the parameter update process, thereby improving the generalization ability and diagnostic accuracy of the water pump fault diagnosis model under different pump types.

[0100] Figure 3This is a flowchart illustrating a water pump fault diagnosis method provided in an embodiment of the present invention. The water pump fault diagnosis method provided in this embodiment is applicable to scenarios where a water pump fault diagnosis model provided in this embodiment is used for water pump fault diagnosis. This water pump fault diagnosis method can be executed by a water pump fault diagnosis device provided in this embodiment, which can be implemented in software and / or hardware. In a specific embodiment, the device can be integrated into an electronic device, such as a computer or server. The following embodiment illustrates the integration of the water pump fault diagnosis device into an electronic device. See reference... Figure 1 The water pump fault diagnosis method in this embodiment may include the following steps:

[0101] Step 301: Obtain the current multidimensional feature data of the target water pump and input the current multidimensional feature data of the target water pump into the fault diagnosis model trained as in any embodiment of the present invention.

[0102] Multidimensional feature current data refers to multi-source signal data collected in real time from the target feedwater pump. A trained fault diagnosis model is a model that has been trained on source domain data and adapted through transfer learning. It includes a feature extractor and a fault classifier, and is used to diagnose faults in the target feedwater pump.

[0103] Specifically, the current data of the multidimensional features of the target pump are prepared and input into the trained fault diagnosis model to prepare for subsequent feature extraction and fault classification.

[0104] Step 302: Use a feature extractor to extract features and obtain the feature vector of the target water pump.

[0105] Specifically, the feature extractor processes the input multidimensional feature data to obtain the feature vector of the target water pump, enabling the fault classifier to diagnose the fault based on these feature vectors.

[0106] Step 303: Use a fault classifier to classify faults and obtain the predicted fault labels for the target feedwater pump.

[0107] Predicted fault labels refer to the fault category or category probability of the current state of the target feedwater pump output by the fault classifier, which is used to determine whether the pump is normal or what type of fault it belongs to.

[0108] Specifically, the feature vector of the target water pump is input into the fault classifier to obtain a predicted fault label for the target pump's state, thereby achieving fault diagnosis.

[0109] In this embodiment, by acquiring the current multidimensional feature data of the target water pump, the current multidimensional feature data of the target water pump is input into the fault diagnosis model trained as in any embodiment of the present invention; feature extraction is performed using a feature extractor to obtain the feature vector of the target water pump; fault classification is performed using a fault classifier to obtain the predicted fault label of the target water pump. Based on the current data of the target pump, key features can be extracted quickly and accurately and fault types can be predicted, thereby improving equipment operation safety and maintenance efficiency.

[0110] Figure 4 This is a schematic diagram of a water pump fault diagnosis model training device provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the device includes:

[0111] The model acquisition module 401 is used to acquire the fault diagnosis model trained using the sample set of the source water pump. The fault diagnosis model includes a feature extractor and a fault classifier.

[0112] The sample acquisition module 402 is used to acquire a sample set of the target water pump, wherein the number of samples in the sample set of the target water pump is less than the number of samples in the sample set of the source water pump.

[0113] The sample training module 403 is used to input the sample set of the target water pump into the feature extractor for feature extraction, obtain the feature vector of each sample of the target water pump, input the feature vector of each sample of the target water pump into the fault classifier for fault classification, and obtain the training output label of each sample of the target water pump.

[0114] The joint feature calculation module 404 is used to calculate the outer product of the feature vector of each sample of the source water pump and the training output label to obtain the first joint feature, and to calculate the outer product of the feature vector of each sample of the target water pump and the training output label to obtain the second joint feature.

[0115] The domain discrimination module 405 is used to input the first joint feature and the second joint feature into the domain discriminator for domain discrimination to obtain the domain discrimination results of all samples;

[0116] The adversarial loss determination module 406 is used to determine the adversarial loss based on the domain discrimination results of all samples and the preset domain labels of all samples;

[0117] The network parameter adjustment module 407 is used to adjust the network parameters of the feature extractor based on adversarial loss.

[0118] In one embodiment, the sample set of the source water pump includes multiple samples of the source water pump, and each sample of the source water pump includes multidimensional feature historical data of the source water pump and a preset fault label.

[0119] In one embodiment, the sample set of the target feedwater pump includes multiple samples of the target feedwater pump. Each sample of the target feedwater pump includes multi-dimensional feature historical data of the target feedwater pump, or includes multi-dimensional feature historical data of the target feedwater pump and a preset fault label.

[0120] In one embodiment, the network parameter adjustment module 407 specifically includes:

[0121] The first loss determination module is used to determine the source domain loss based on the training output labels of all samples from the source water pump and the preset fault labels of all samples.

[0122] The second loss determination module is used to determine the target domain loss based on the preset fault labels of all samples of the target water pump;

[0123] The overall loss determination module is used to determine the overall loss based on the source domain loss, target domain loss, and adversarial loss.

[0124] The parameter tuning module is used to adjust the network parameters of the feature extractor based on the overall loss.

[0125] In one embodiment, the parameter adjustment module is specifically used for:

[0126] The overall gradient passed to the feature extractor is determined based on the overall loss. The overall gradient includes the source domain gradient, the target domain gradient, and the adversarial gradient.

[0127] Multiply the adversarial gradient by a preset negative coefficient to obtain the inverted adversarial gradient;

[0128] The network parameters of the feature extractor are adjusted based on the source domain gradient, the target domain gradient, and the inverted adversarial gradient.

[0129] In one embodiment, the multidimensional feature historical data is obtained through the following method:

[0130] Using a preset sliding window with a preset step size, the signal is obtained by sliding the signal from the historical signal data to obtain multiple historical window signals. The historical signal data includes historical high-frequency signals and historical low-frequency signals.

[0131] Calculate the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal, and then concatenate the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal to obtain multiple multidimensional feature historical data.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is merely an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the functional modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0133] The apparatus of this invention acquires a fault diagnosis model trained using a sample set of source feedwater pumps. The fault diagnosis model includes a feature extractor and a fault classifier, providing a foundational model for subsequent target domain fault diagnosis. It acquires a sample set of target feedwater pumps, where the number of samples is less than the number of samples in the source feedwater pump sample set, thus providing a limited sample set for target feedwater pumps and a data foundation for transfer learning. The sample set of target feedwater pumps is input into the feature extractor for feature extraction, obtaining a feature vector for each sample of the target feedwater pump. The feature vector of each sample of the target feedwater pump is then input into the fault classifier for fault classification, resulting in a training output label for each sample of the target feedwater pump. This allows for the extraction of effective features from the limited sample set of the target feedwater pump and the generation of preliminary prediction results, providing a basis for subsequent domain adaptation and model optimization. The apparatus also calculates the values ​​for each sample of the source feedwater pump. The first joint feature is obtained by multiplying the feature vector of each sample of the target water pump by the outer product of the feature vector and the training output label. The second joint feature is obtained by multiplying the feature vector of each sample by the outer product of the training output label. This allows for the joint representation of feature vectors and prediction information, enhancing the discriminability of samples in the domain discriminator and providing effective input for adversarial training. The first and second joint features are then input into the domain discriminator for domain discrimination, yielding the domain discrimination results for all samples. This provides a basis for determining the adversarial loss. The adversarial loss is determined based on the domain discrimination results of all samples and the preset domain labels of all samples, providing optimization direction for the feature extractor and enhancing the model's generalization ability in the target domain. Adjusting the network parameters of the feature extractor based on the adversarial loss makes it difficult for the feature vectors extracted by the feature extractor to distinguish the sample source, reducing the feature difference between the source and target domains, thereby improving the accuracy and stability of fault diagnosis of the target water pump.

[0134] Figure 5 This is a schematic diagram of a water pump fault diagnosis device provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the device includes:

[0135] The data acquisition module 501 is used to acquire the current data of the multidimensional features of the target water pump and input the current data of the multidimensional features of the target water pump into the fault diagnosis model trained as in any embodiment of the present invention.

[0136] The feature vector extraction module 502 is used to extract features using a feature extractor to obtain the feature vector of the target water pump.

[0137] The fault classification module 503 is used to classify faults using a fault classifier to obtain the predicted fault label of the target feedwater pump.

[0138] The apparatus of this invention acquires the current multidimensional feature data of the target water pump, inputs the current multidimensional feature data of the target water pump into a fault diagnosis model trained as in any embodiment of this invention, extracts features using a feature extractor to obtain the feature vector of the target water pump, and classifies faults using a fault classifier to obtain the predicted fault label of the target water pump. Based on the current data of the target pump, it can quickly and accurately extract key features and predict fault types, thereby improving equipment operation safety and maintenance efficiency.

[0139] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a computer system 600 suitable for implementing an electronic device according to embodiments of the present invention. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0140] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer system 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0141] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube, liquid crystal display, etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a modem, etc. Communication section 609 performs communication processing via a network such as the Internet. Drive 610 is also connected to I / O interface 605 as needed. Removable media 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 610 as needed so that computer programs read from them can be installed into storage section 608 as needed.

[0142] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by central processing unit (CPU) 601, it performs the functions defined above in the system of this invention.

[0143] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, etc., or any suitable combination thereof.

[0144] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0145] The modules and / or units described in the embodiments of this invention can be implemented in software or hardware. The described modules and / or units can also be housed in a processor. For example, a processor may be described as including a model acquisition module, a sample acquisition module, a sample training module, a joint feature calculation module, a domain discrimination module, an adversarial loss determination module, and a network parameter adjustment module; or, a processor may be described as including a data acquisition module, a feature vector extraction module, and a fault classification module. The names of these modules do not necessarily constitute a limitation on the module itself.

[0146] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to include:

[0147] A fault diagnosis model trained using a sample set from the source water pump is obtained. This model includes a feature extractor and a fault classifier. A sample set from the target water pump is obtained, with fewer samples than those from the source water pump. The target water pump sample set is input into the feature extractor for feature extraction, yielding a feature vector for each sample. This feature vector is then input into the fault classifier for fault classification, resulting in a training output label for each sample. The outer product of the feature vector and training output label for each sample from the source water pump is calculated to obtain a first joint feature. The outer product of the feature vector and training output label for each sample from the target water pump is also calculated to obtain a second joint feature. The first and second joint features are input into a domain discriminator for domain discrimination, yielding the domain discrimination results for all samples. An adversarial loss is determined based on the domain discrimination results and preset domain labels for all samples. The network parameters of the feature extractor are adjusted based on the adversarial loss.

[0148] The technical solution of this invention involves obtaining a fault diagnosis model trained using a sample set of source feedwater pumps. This model includes a feature extractor and a fault classifier, providing a foundational model for subsequent target domain fault diagnosis. A sample set of target feedwater pumps is obtained, with the number of samples in the target feedwater pump sample set being less than the number of samples in the source feedwater pump sample set, thus providing a limited sample set for transfer learning. The sample set of target feedwater pumps is input into the feature extractor for feature extraction, yielding a feature vector for each sample of the target feedwater pump. This feature vector is then input into the fault classifier for fault classification, resulting in a training output label for each sample of the target feedwater pump. This process extracts effective features from the limited sample set of the target feedwater pump and generates preliminary prediction results, providing a basis for subsequent domain adaptation and model optimization. The calculation of each sample from the source feedwater pump... The first joint feature is obtained by multiplying the feature vector of the target water pump by the outer product of the training output label and the feature vector of each sample. The second joint feature is obtained by multiplying the feature vector of each sample by the training output label. This allows for the joint representation of feature vectors and prediction information, enhancing the discriminability of samples in the domain discriminator and providing effective input for adversarial training. The first and second joint features are then input into the domain discriminator for domain discrimination, yielding the domain discrimination results for all samples. This provides a basis for determining the adversarial loss. The adversarial loss is determined based on the domain discrimination results of all samples and the preset domain labels of all samples, providing optimization direction for the feature extractor and enhancing the model's generalization ability in the target domain. Adjusting the network parameters of the feature extractor based on the adversarial loss makes it difficult for the feature vectors extracted by the feature extractor to distinguish the sample source, reducing the feature differences between the source and target domains, thereby improving the accuracy and stability of fault diagnosis for the target water pump.

[0149] Alternatively, the aforementioned computer-readable medium carries one or more programs that, when executed by a device, cause the device to include:

[0150] The current multidimensional feature data of the target water pump is obtained, and the current multidimensional feature data of the target water pump is input into the fault diagnosis model trained as in any embodiment of the present invention; the feature extractor is used to extract features to obtain the feature vector of the target water pump; the fault classifier is used to classify the fault to obtain the predicted fault label of the target water pump.

[0151] The technical solution of this invention obtains the current multi-dimensional feature data of the target water pump, inputs the current multi-dimensional feature data of the target water pump into the fault diagnosis model trained as in any embodiment of this invention, extracts features using a feature extractor to obtain the feature vector of the target water pump, and classifies faults using a fault classifier to obtain the predicted fault label of the target water pump. Based on the current data of the target pump, it can quickly and accurately extract key features and predict fault types, thereby improving equipment operation safety and maintenance efficiency.

[0152] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the water pump fault diagnosis model training method or the water pump fault diagnosis method provided in any embodiment of this invention.

[0153] In the implementation of a computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​as well as conventional procedural programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0154] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0155] It should be noted that the collection, use, storage, sharing, and transfer of user personal information involved in the technical solution of this invention all comply with the provisions of relevant laws and regulations, and require notification to the user and obtaining the user's consent or authorization. Where applicable, user personal information has undergone de-identification and / or anonymization and / or encryption technical processing. In addition, a corresponding operation entry is provided for the user to choose to agree to or reject the automated decision result; if the user chooses to reject, the process proceeds to the expert decision-making process.

[0156] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for training a fault diagnosis model for a water pump, characterized in that, include: A fault diagnosis model is obtained by training a sample set of the source water pump, the fault diagnosis model including a feature extractor and a fault classifier; Obtain a sample set of the target water pump, wherein the number of samples in the sample set of the target water pump is less than the number of samples in the sample set of the source water pump; The sample set of the target water pump is input into the feature extractor for feature extraction to obtain the feature vector of each sample of the target water pump. The feature vector of each sample of the target water pump is input into the fault classifier for fault classification to obtain the training output label of each sample of the target water pump. The first joint feature is obtained by calculating the outer product of the feature vector and the training output label of each sample of the source water pump; the second joint feature is obtained by calculating the outer product of the feature vector and the training output label of each sample of the target water pump. The first joint feature and the second joint feature are input into the domain discriminator for domain discrimination to obtain the domain discrimination results of all samples; The adversarial loss is determined based on the domain discrimination results of all samples and the preset domain labels of all samples; The network parameters of the feature extractor are adjusted based on the adversarial loss.

2. The method according to claim 1, characterized in that, The sample set of the source water pump includes multiple samples of the source water pump, and each sample of the source water pump includes multi-dimensional feature historical data and preset fault labels of the source water pump.

3. The method according to claim 2, characterized in that, The sample set of the target water pump includes multiple samples of the target water pump. Each sample of the target water pump includes the multidimensional feature historical data of the target water pump, or includes the multidimensional feature historical data of the target water pump and a preset fault label.

4. The method according to claim 3, characterized in that, Adjusting the network parameters of the feature extractor based on the adversarial loss includes: The source domain loss is determined based on the training output labels of all samples from the source water pump and the preset fault labels of all samples. The target domain loss is determined based on the preset fault labels of all samples of the target water pump; The overall loss is determined based on the source domain loss, the target domain loss, and the adversarial loss; The network parameters of the feature extractor are adjusted based on the overall loss.

5. The method according to claim 4, characterized in that, Adjusting the network parameters of the feature extractor based on the overall loss includes: The overall gradient passed to the feature extractor is determined based on the overall loss, and the overall gradient includes the source domain gradient, the target domain gradient, and the adversarial gradient. Multiply the adversarial gradient by a preset negative coefficient to obtain the inverted adversarial gradient; The network parameters of the feature extractor are adjusted based on the source domain gradient, the target domain gradient, and the inverted adversarial gradient.

6. The method according to claim 2 or 3, characterized in that, The multidimensional feature historical data is obtained through the following method: Using a preset sliding window with a preset step size, multiple historical window signals are obtained by sliding and extracting signals from historical signal data. The historical signal data includes historical high-frequency signals and historical low-frequency signals. Calculate the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal, and then concatenate the statistical characteristics of historical high-frequency signals and the mean characteristics of historical low-frequency signals in each historical window signal to obtain multiple multidimensional feature historical data.

7. A method for diagnosing water pump faults, characterized in that, include: Obtain the current data of the multidimensional features of the target water pump, and input the current data of the multidimensional features of the target water pump into the fault diagnosis model trained by any one of claims 1 to 6; The feature extractor is used to extract features to obtain the feature vector of the target water pump. The fault classifier is used to classify faults and obtain the predicted fault labels of the target feedwater pump.

8. 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 program to implement the water pump fault diagnosis model training method as described in any one of claims 1 to 6, or the processor executes the program to implement the water pump fault diagnosis method as described in claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the water pump fault diagnosis model training method as described in any one of claims 1 to 6, or when the program is executed by the processor, it implements the water pump fault diagnosis method as described in claim 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the water pump fault diagnosis model training method as described in any one of claims 1 to 6, or the computer program that, when executed by a processor, implements the water pump fault diagnosis method as described in claim 7.

Images