Cylinder cover fault diagnosis method, device and system

Abstract

The invention discloses a cylinder cover fault diagnosis method, device and system, and the method comprises the steps: carrying out the interference signal elimination operation of a vibration signal collected by a vibration sensor in the operation process of an internal combustion engine unit, and obtaining a to-be-detected cylinder cover vibration signal; and calling the cylinder cover fault diagnosis model to process the signal features extracted from the to-be-detected cylinder cover vibration signals to obtain a preliminary cylinder cover diagnosis result representing whether the signal features have faults or not. Under the condition that the preliminary cylinder cover diagnosis result represents that a fault exists, similarity matching is carried out on the signal feature and a fault signal feature corresponding to each pre-stored fault type, and a target fault type is obtained; the fault degree of the cylinder cover is determined according to the dispersion degree of the identified vibration signal of the cylinder cover to be detected; and generating a cylinder cover diagnosis report based on the preliminary cylinder cover diagnosis result, the target fault type and the fault degree, and outputting the cylinder cover diagnosis report. Based on the above process, the accuracy and efficiency of cylinder cover fault diagnosis based on the cylinder cover vibration signal can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of mechanical equipment status monitoring and fault diagnosis, and more specifically, to a cylinder head fault diagnosis method, device and system. Background Art

[0002] Diesel generator sets, gasoline generator sets and various internal combustion engine sets all have cylinder heads, which play a vital role in these sets. For example, through the close cooperation between the cylinder head and the cylinder body, a closed combustion environment is formed to jointly realize the combustion of fuel and the output of power. Therefore, the operating status of the cylinder head has a significant impact on the performance of the set.

[0003] Taking diesel engine units as an example, the current detection and diagnosis of diesel engine cylinder head status mainly relies on manual regular inspection and experience judgment. Maintenance personnel make a preliminary judgment on whether the cylinder head is abnormal by observing the cylinder head surface and listening to vibration noise. When a problem is found, the engine is stopped for inspection. The fault detection results under this detection method rely on human experience and subjective judgment, which is prone to misdiagnosis and missed diagnosis. In addition, the maintenance personnel have a long cycle for cylinder head inspection, which causes the diesel engine to run for a long time with cylinder head failure, affecting the performance of the unit.

[0004] Therefore, how to improve the accuracy and efficiency of cylinder head fault detection has become an urgent problem to be solved. Summary of the invention

[0005] In view of the above problems, the present application provides a cylinder head fault diagnosis method, device and system to achieve the purpose of improving the accuracy and efficiency of cylinder head fault detection. The specific scheme is as follows:

[0006] In a first aspect, the present application provides a cylinder head fault diagnosis method, which is applied to a processor of a cylinder head fault diagnosis system, wherein the cylinder head fault diagnosis system at least further comprises: when collecting a vibration signal, a vibration sensor disposed on the cylinder head, and the cylinder head fault diagnosis method comprises:

[0007] During the operation of the internal combustion engine unit, obtaining a vibration signal at a current time point collected by the vibration sensor;

[0008] Eliminating interference signals in the vibration signal to obtain a cylinder head vibration signal to be detected;

[0009] Extracting signal features from the cylinder head vibration signal to be detected;

[0010] Calling a pre-trained cylinder head fault diagnosis model, processing the signal features, and obtaining a preliminary cylinder head diagnosis result, wherein the cylinder head fault diagnosis model uses signal feature samples of cylinder head vibration signal samples as training samples, the training samples are correspondingly marked with sample labels, and the training is performed using the sample labels as output, wherein the sample labels are labels that characterize whether the signal feature samples have a fault;

[0011] In the case where the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault, the signal feature is matched with a pre-stored fault signal feature corresponding to each fault type to obtain a target fault signal feature whose similarity with the signal feature reaches a preset threshold, and the fault type corresponding to the target fault signal feature is determined as the target fault type;

[0012] Determining the degree of failure of the cylinder head according to the identified discrete degree of the cylinder head vibration signal to be detected;

[0013] A cylinder head diagnosis report of the internal combustion engine group is generated and outputted based on the preliminary cylinder head diagnosis result, the target fault type and the fault degree.

[0014] In a possible implementation, after extracting the signal feature in the cylinder head vibration signal to be detected, the method further includes:

[0015] Obtaining signal characteristics corresponding to a preset number of time points before the current time point;

[0016] Generate a signal feature sequence to be predicted based on signal features corresponding to a preset number of time points before the current time point and the signal features corresponding to the current time point;

[0017] Calling a pre-trained cylinder head vibration signal prediction model, performing prediction processing on the signal feature sequence to be predicted, and obtaining a signal feature prediction sequence, wherein the signal feature prediction sequence includes: a signal feature of a cylinder head vibration signal at at least one prediction time point, and the cylinder head vibration signal prediction model is a model obtained by training with the signal feature sequence of the historical cylinder head vibration signals in the previous order as feature values ​​and the signal feature sequence of the historical cylinder head vibration signals in the subsequent order as target values ​​in the historical cylinder head vibration signals sorted according to the time series;

[0018] The pre-trained cylinder head fault diagnosis model is called to process the signal features corresponding to each prediction time point in the signal feature prediction sequence respectively to obtain a preliminary prediction diagnosis result corresponding to each prediction time point.

[0019] In a possible implementation, the training process of the cylinder head vibration signal prediction model includes:

[0020] Acquire at least one set of historical cylinder head vibration signals;

[0021] Extracting the signal feature at each time point in the historical cylinder head vibration signal to obtain at least one historical signal feature sequence;

[0022] Dividing the historical signal feature sequence into a preceding feature sequence and a succeeding feature sequence, wherein the time point of each signal feature in the preceding feature sequence is earlier than the time point of each signal feature in the succeeding feature sequence;

[0023] Based on the preceding feature sequence and the succeeding feature sequence, a prediction model to be trained is trained to obtain a signal prediction model, wherein the training process uses the preceding feature sequence as a feature value and uses the succeeding feature sequence corresponding to the preceding feature sequence as a target value for training;

[0024] Calling the signal prediction model to perform prediction processing on any of the preceding feature sequences to obtain a predicted feature sequence;

[0025] Determine whether the error between the subsequent feature sequence corresponding to the preceding feature sequence and the predicted feature sequence is greater than a preset error;

[0026] If the error between the subsequent feature sequence and the predicted feature sequence is not greater than the preset error, determining that the signal prediction model is a trained cylinder head vibration signal prediction model;

[0027] If the error between the subsequent feature sequence and the predicted feature sequence is greater than the preset error, the model parameters of the signal prediction model are adjusted to obtain the prediction model to be trained, and the step of training the prediction model to be trained based on the preceding feature sequence and the subsequent feature sequence to obtain the signal prediction model is returned to execute until the error between the subsequent feature sequence and the predicted feature sequence is no greater than the preset error.

[0028] In a possible implementation, the training process of the cylinder head fault diagnosis model includes:

[0029] Acquire at least one set of cylinder head vibration signal samples and at least one fault signal feature corresponding to a fault type;

[0030] Extracting the signal feature at each time point in the cylinder head vibration signal sample to obtain a signal feature set;

[0031] Based on the fault signal feature corresponding to the fault type, determining the sample label corresponding to each signal feature in the feature set, the sample label including: a label indicating whether there is a fault;

[0032] Based on the feature set and the sample labels corresponding to each signal feature in the feature set, the fault diagnosis model to be trained is trained to obtain a cylinder head fault diagnosis model, wherein the training process includes: using the signal feature samples of the cylinder head vibration signal samples as feature values, and using the sample labels corresponding to the signal feature samples as target values ​​for training.

[0033] In a possible implementation, removing the interference signal in the vibration signal to obtain the cylinder head vibration signal to be detected includes:

[0034] Calling a filter to filter out noise within a preset frequency range in the vibration signal to obtain a first vibration signal;

[0035] Eliminate data points that meet preset elimination conditions in the first vibration signal to obtain a second vibration signal, wherein the preset elimination conditions at least include: an amplitude greater than a preset amplitude threshold, and a time interval between data points less than a preset interval threshold;

[0036] The vibration signal of the cylinder head is separated from the second vibration signal to obtain the cylinder head vibration signal to be detected.

[0037] In a possible implementation, separating the vibration signal of the cylinder head from the second vibration signal to obtain the cylinder head vibration signal to be detected includes:

[0038] Segmenting the second vibration signal according to a preset segment duration to obtain a plurality of segment signals;

[0039] Centering and whitening each of the segmented signals respectively to obtain a segmented signal to be analyzed corresponding to each of the segmented signals;

[0040] Invoke an independent component analysis algorithm to analyze and process each of the segmented signals to be analyzed to obtain a separation matrix;

[0041] According to the timing relationship of the time ranges corresponding to each of the segmented signals, determining the segmented signal with the earliest timing in the time range as the target segmented signal;

[0042] The product of the separation matrix and the target segmented signal is used as the cylinder head vibration signal to be detected.

[0043] In a possible implementation, the method further includes:

[0044] Determining a corresponding fault warning method according to the fault type and the fault degree described in the cylinder head diagnostic report;

[0045] Based on the fault warning method, information indicating that a fault of the fault type exists in the cylinder head of the internal combustion engine unit is sent to the front end.

[0046] In a possible implementation, the method further includes:

[0047] When the preliminary cylinder head diagnosis result is a label indicating that the cylinder head is faulty, a target repair solution corresponding to the target fault type and the fault degree is matched from a pre-established fault repair solution repository and outputted.

[0048] A second aspect of the present application provides a cylinder head fault diagnosis device, which is applied to a processor of a cylinder head fault diagnosis system, wherein the cylinder head fault diagnosis system at least further comprises: a vibration sensor disposed on the cylinder head when collecting vibration signals, and the cylinder head fault diagnosis device comprises:

[0049] A signal acquisition unit, used to acquire the vibration signal at a current time point collected by the vibration sensor during the operation of the internal combustion engine unit;

[0050] A signal elimination unit, used for eliminating interference signals in the vibration signal to obtain a cylinder head vibration signal to be detected;

[0051] A feature extraction unit, used to extract signal features from the cylinder head vibration signal to be detected;

[0052] a model processing unit, used to call a pre-trained cylinder head fault diagnosis model, process the signal features, and obtain a preliminary cylinder head diagnosis result, wherein the cylinder head fault diagnosis model uses a signal feature sample of a cylinder head vibration signal sample as a training sample, the training sample is correspondingly marked with a sample label, and the training is performed using the sample label as an output, wherein the sample label is a label that characterizes whether the signal feature sample has a fault;

[0053] a type determination unit, configured to, when the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault, perform similarity matching between the signal feature and a fault signal feature corresponding to each pre-stored fault type, obtain a target fault signal feature whose similarity with the signal feature reaches a preset threshold, and determine the fault type corresponding to the target fault signal feature as a target fault type;

[0054] a degree determination unit, configured to determine the degree of failure of the cylinder head according to the identified discrete degree of the vibration signal of the cylinder head to be detected;

[0055] The result output unit is used to generate and output a cylinder head diagnosis report of the internal combustion engine unit based on the preliminary cylinder head diagnosis result, the target fault type and the fault degree.

[0056] A third aspect of the present application provides a cylinder head fault diagnosis system, comprising:

[0057] When collecting vibration signals, a vibration sensor is provided on the cylinder head;

[0058] A processor is used to implement any step in the cylinder head fault diagnosis method.

[0059] It can be seen from the above technical solutions that the cylinder head fault diagnosis method provided by the embodiment of the present application performs interference signal elimination operation on the vibration signal collected by the vibration sensor to obtain the cylinder head vibration signal used for subsequent fault diagnosis, thereby avoiding vibration interference with adjacent components of the cylinder head, such as valves, camshafts, etc. Further, the pre-trained cylinder head fault diagnosis model is used to perform quantitative fault state diagnosis on the signal characteristics of the cylinder head vibration signal, which can improve the objectivity of fault diagnosis and the accuracy and precision of cylinder head fault diagnosis based on the cylinder head vibration signal compared to human subjective judgment.

[0060] In addition, when the preliminary diagnostic results indicate that there is a fault in the cylinder head, the signal characteristics of the cylinder head are further matched with the fault signal characteristics of each fault type for similarity and the degree of discreteness is identified to achieve quantitative verification of the fault type and degree of the cylinder head, avoiding the complex manual feature analysis process in the traditional method, greatly simplifying the signal processing process, and significantly improving the diagnostic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0062] Figure 1 A schematic diagram of a flow chart of a cylinder head fault diagnosis method provided in an embodiment of the present application;

[0063] Figure 2 A schematic diagram of the structure of a cylinder head vibration signal diagnosis model based on CNN provided in an embodiment of the present application;

[0064] Figure 3 A schematic structural diagram of a cylinder head fault diagnosis device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0066] At present, vibration signal analysis technology has been widely used in the field of mechanical equipment status monitoring and fault diagnosis. For example, the diagnosis of faults of different components is achieved by collecting and processing the vibration signals of the unit. However, due to the complex structure of the cylinder head, variable working conditions, and various fault modes, its vibration signal presents complex characteristics of non-stationary and nonlinearity. In addition, the vibration of the cylinder head is easily disturbed by the vibration signals of adjacent components such as valves and camshafts, making it difficult for the collected vibration signal to accurately reflect the state of the cylinder head itself, thereby reducing the accuracy of the cylinder head fault diagnosis.

[0067] Traditional diagnostic methods rely on manual experience, cannot standardize the diagnostic results of cylinder heads, and have low diagnostic efficiency. Traditional diagnostic models mostly use manual feature extraction and shallow learning classification algorithms. Due to their limited expression ability and fault resolution ability, insufficient diagnostic generalization performance, and difficulty in adapting to the state recognition requirements of cylinder heads under complex working conditions, the accuracy of cylinder head fault diagnosis is low.

[0068] In order to solve the above problems, an embodiment of the present application provides a cylinder head fault diagnosis method for eliminating interference signals, and uses a cylinder head fault diagnosis model obtained through deep learning training to perform fault diagnosis on relatively interference-free cylinder head vibration signals, thereby improving the accuracy and efficiency of diagnosis and reducing reliance on manual experience.

[0069] Reference Figure 1 , a flow chart of a cylinder head fault diagnosis method provided in an embodiment of the present application specifically illustrates a cylinder head fault diagnosis method.

[0070] Step S110: during the operation of the internal combustion engine unit, a vibration signal at a current time point collected by a vibration sensor is obtained.

[0071] The cylinder head fault diagnosis method of the embodiment of the present application is applied to a processor of a cylinder head fault diagnosis system, and the cylinder head fault diagnosis system at least further includes: a vibration sensor disposed on the cylinder head when collecting vibration signals.

[0072] The main function of the vibration sensor is to receive the mechanical quantity of the cylinder head vibration during the operation of the internal combustion engine unit and convert it into an electrical signal proportional to it, which is the vibration signal. Optionally, the vibration sensor can use an acceleration sensor with a measurement frequency range of 0.1Hz~10kHz and a sensitivity of 100mV / g. The vibration sensor can be installed in different positions of the cylinder head, such as the valve chamber, combustion chamber, etc., to fully collect the vibration signal of the cylinder head.

[0073] Step S120, eliminating interference signals in the vibration signal to obtain the cylinder head vibration signal to be detected.

[0074] It is understandable that when the vibration sensor collects the vibration signal of the cylinder head, the cylinder head is affected by the vibration of adjacent components, so that the collected vibration signal is mixed with vibration signals of other components and interference signals such as noise.

[0075] Therefore, before diagnosing the vibration signal, the embodiment of the present application pre-processes the vibration signal to remove the interference signal therein, obtains a relatively interference-free cylinder head vibration signal, and improves the accuracy of subsequent diagnosis. Optionally, in the signal processing process of the processor, an algorithm such as digital filtering or ramp function can be used to filter the interference signal, or the source and propagation path of the interference signal in the vibration signal can be analyzed, and the interference source can be located and eliminated in the process of collecting the signal, thereby fundamentally reducing the impact of the interference signal on the vibration signal.

[0076] Step S130: extracting signal features from the cylinder head vibration signal to be detected.

[0077] Among them, the signal features may include peak features, phase features, frequency features, etc., and the signal features in the cylinder head vibration signal to be detected are calculated according to the feature extraction algorithms or formulas corresponding to different signal features. Alternatively, the cylinder head vibration signal to be detected is input into a pre-trained signal feature recognition model, and the signal features are recognized using the model.

[0078] In the embodiment of the present application, the formula for extracting the amplitude characteristic of the cylinder head vibration signal to be detected is as follows: rms , peak value X peak Kurtosis .

[0079] (1)

[0080] Among them, x c (t i ) represents the cylinder head vibration signal x to be detected c (t) is the value of the i-th data point, and N is the total number of data points in the cylinder head vibration signal to be detected.

[0081] When extracting the phase characteristics of the cylinder head vibration signal to be detected, firstly c (t) Perform Hilbert transform to obtain the analytical signal ; Further, take The phase angle ,statistics In different phase ranges The probability distribution within , and used as the phase feature of the cylinder head vibration signal to be detected.

[0082] Extract the cylinder head vibration signal x to be detected cThe process of frequency characteristics of (t) includes: c (t) Perform fast Fourier transform to obtain the spectrum , take the spectrum The center of gravity frequency f c and bandwidth B as x c (t) frequency characteristics. Among them, the center of gravity frequency f is calculated c The formula for the sum of bandwidth B is as follows (2):

[0083] (2)

[0084] Among them, f0 is the spectrum resolution and K is the number of spectrum lines.

[0085] Optionally, the embodiment of the present application designs a feature extraction algorithm suitable for non-stationary, nonlinear and complex signals based on the characteristics of the cylinder head vibration signal, thereby greatly improving the diagnostic accuracy of the cylinder head fault.

[0086] Step S140, calling the pre-trained cylinder head fault diagnosis model, processing the signal features, and obtaining a preliminary cylinder head diagnosis result.

[0087] The cylinder head fault diagnosis model uses the signal feature samples of the cylinder head vibration signal samples as training samples. The training samples are marked with sample labels and trained with sample labels as output. The sample labels are labels that characterize whether the signal feature samples have faults. Based on this, the cylinder head fault diagnosis model can confirm whether it belongs to the signal features of the fault according to the signal features input into the model, and output the corresponding labels as the preliminary cylinder head diagnosis results.

[0088] Optionally, the cylinder head fault diagnosis model may use a machine learning model such as an SVM model, a convolutional neural network (CNN), or a random forest, wherein the process of training the machine learning model to obtain the cylinder head fault diagnosis model may include:

[0089] Acquire at least one group of cylinder head vibration signal samples and at least one fault signal feature corresponding to a fault type; extract the signal features at each time point in the cylinder head vibration signal samples to obtain a signal feature set; determine the sample label corresponding to each of the signal features in the feature set based on the fault signal features corresponding to the fault type, wherein the sample label includes: a label characterizing whether a fault exists; train a fault diagnosis model to be trained based on the feature set and the sample label corresponding to each signal feature in the feature set to obtain a cylinder head fault diagnosis model, wherein the training process includes: using the signal feature samples of the cylinder head vibration signal samples as feature values ​​and using the sample labels corresponding to the signal feature samples as target values ​​for training.

[0090] Take the training process of SVM model as an example to illustrate, select n groups of cylinder head vibration signal samples According to steps S120-130, each cylinder head vibration signal sample is processed to obtain a signal feature sample Fi of each cylinder head vibration signal sample, and a signal feature set Fi can be obtained. According to the known fault types and the fault signal characteristics of each fault type, a sample label is marked for each signal feature in the feature set to obtain the label set corresponding to the feature set. , where yi=1 means that the i-th signal feature in the feature set is normal, and yi=-1 means that the i-th signal feature in the feature set is faulty.

[0091] Select RBF kernel function , build the SVM model, and transform the feature set As the input of the SVM model, the label set As the output of the SVM model, it is trained. And the SVM model is optimized according to the following formula (3). By solving formula (3), the optimal value suitable for the SVM model is obtained. .

[0092] (3)

[0093] in, is the Lagrange multiplier, is the penalty factor.

[0094] Furthermore, the classification decision function of the SVM model, that is, the output function, is determined as follows (4):

[0095] (4)

[0096] in, is the bias term. , then the preliminary diagnosis result is normal. , the preliminary diagnosis result is a fault.

[0097] In another possible implementation, a cylinder head fault diagnosis model can also be constructed based on a convolutional neural network (CNN), referring to Figure 2 , a schematic diagram of a cylinder head vibration signal diagnosis model based on CNN provided in an embodiment of the present application, wherein the cylinder head vibration signal to be detected belonging to the time domain vibration signal is first The local features of the time-frequency image I are then extracted through the convolution layer. The pooling layer compresses the feature size. The fully connected layer maps the features to the fault type. The Softmax layer outputs the probability of the features in each fault type. The fault type corresponding to the maximum probability value is taken as The training of the cylinder head vibration signal diagnosis model based on CNN is similar to the training process of the SVM model mentioned above, which will not be repeated here.

[0098] Step S150, when the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault, the signal feature is matched with the fault signal feature corresponding to each pre-stored fault type for similarity, and a target fault signal feature whose similarity with the signal feature reaches a preset threshold is obtained, and the fault type corresponding to the target fault signal feature is determined as the target fault type.

[0099] Step S160, determining the fault degree of the cylinder head according to the discrete degree of the identified cylinder head vibration signal to be detected.

[0100] As can be seen from the above, the preliminary cylinder head diagnosis result of step S140 can only determine whether the current cylinder head has a fault, but the fault type and fault degree cannot be determined. Therefore, in the embodiment of the present application, when the preliminary cylinder head diagnosis result is that the cylinder head has a fault, the cylinder head vibration signal to be detected is further analyzed and processed to determine the fault type and fault degree, thereby improving the accuracy of the cylinder head fault diagnosis.

[0101] The processor pre-stores the fault signal features of each fault type. The specific fault type is determined based on the feature similarity between the signal features of the cylinder head vibration signal to be detected and the fault signal features of each fault type. The similarity can be measured using Euclidean distance, Mahalanobis distance, etc. The size relationship of the feature similarity between the signal features and the fault signal features of each fault type is compared, and the fault type with the highest similarity is used as the target fault type.

[0102] Further, the discreteness of the cylinder head vibration signal to be detected is identified, for example, by calculating the variance, standard deviation, etc. of the cylinder head vibration signal to be detected, and the obtained discreteness is used as the fault degree reflected by the cylinder head vibration signal to be detected.

[0103] In the embodiment of the present application, the Mahalanobis distance is used to determine the discrete degree of the cylinder head vibration signal to be detected. The Mahalanobis distance is a method for measuring the distance between two points. It takes into account the covariance relationship between data, so it can more accurately reflect the real distance of data in multidimensional space. In signal processing, the Mahalanobis distance can be used to evaluate the discrete degree of the signal. The embodiment of the present application normalizes the calculation result of the Mahalanobis distance to the interval [0, 1]. Therefore, if the discrete degree of the cylinder head vibration signal to be detected determined by the Mahalanobis distance is less than 0.3, the fault degree can be determined to be a mild fault; if the discrete degree is in the range of 0.3~0.7, the fault degree is determined to be; if the discrete degree is greater than 0.7, the fault degree is determined to be. Optionally, "mild fault", "moderate fault" and "severe fault" can be output as the fault degree, or the discrete degree value can be output as the fault degree.

[0104] Step S170: Generate and output a cylinder head diagnosis report of the internal combustion engine unit based on the preliminary cylinder head diagnosis result, the target fault type and the fault degree.

[0105] Select the output method according to the report output requirements in the unit diagnostic system. If the unit diagnostic system includes a display device for displaying data, the cylinder head diagnostic report can be output to the display device first so that the personnel who monitor the equipment can check it in time. If the unit diagnostic system requires remote monitoring, the cylinder head diagnostic report can be sent to the staff by email, phone, SMS, etc. according to the contact information pre-filled by the staff.

[0106] In summary, the cylinder head fault diagnosis method provided by the embodiment of the present application performs interference signal elimination operation on the vibration signal collected by the vibration sensor to obtain the cylinder head vibration signal used for subsequent fault diagnosis, thereby avoiding vibration interference with adjacent components of the cylinder head, such as valves, camshafts, etc. Further, the pre-trained cylinder head fault diagnosis model is used to perform quantitative fault state diagnosis on the signal characteristics of the cylinder head vibration signal, which can improve the objectivity of fault diagnosis and the accuracy and precision of cylinder head fault diagnosis based on the cylinder head vibration signal compared to human subjective judgment.

[0107] In addition, when the preliminary diagnostic results indicate that there is a fault in the cylinder head, the signal characteristics of the cylinder head are further matched with the fault signal characteristics of each fault type for similarity and the degree of discreteness is identified to achieve quantitative verification of the fault type and degree of the cylinder head, avoiding the complex manual feature analysis process in the traditional method, greatly simplifying the signal processing process, and significantly improving the diagnostic efficiency.

[0108] Next, other possible implementations of the above cylinder head fault diagnosis method are described.

[0109] In one possible implementation, the cylinder head fault diagnosis method also includes: obtaining signal features corresponding to a preset number of time points before the current time point; generating a signal feature sequence to be predicted based on the signal features corresponding to the preset number of time points before the current time point and the signal features corresponding to the current time point; calling a pre-trained cylinder head vibration signal prediction model to perform prediction processing on the signal feature sequence to be predicted to obtain a signal feature prediction sequence, the signal feature prediction sequence including: signal features of a cylinder head vibration signal at at least one prediction time point, the cylinder head vibration signal prediction model is a model obtained by training with the signal feature sequence of the preceding historical cylinder head vibration signals sorted according to the time series as feature values ​​and the signal feature sequence of the succeeding historical cylinder head vibration signals as target values; calling a pre-trained cylinder head fault diagnosis model to process the signal features corresponding to each prediction time point in the signal feature prediction sequence respectively to obtain a preliminary prediction diagnosis result corresponding to each prediction time point.

[0110] In an embodiment of the present application, a cylinder head vibration signal at the current moment and before the current moment is acquired, the future development trend and change law of the cylinder head vibration signal are predicted, and based on the predicted vibration signal, it is determined whether there is a possibility of a fault.

[0111] Specifically, the embodiment of the present application can use only the signal characteristics at the current moment as the prediction basis and input them into the cylinder head vibration signal prediction model for prediction, or obtain the signal characteristics of the time point before the current moment as the model input for prediction. It can be understood that the more signal characteristics used as the prediction basis, the more accurate the prediction result. The preset number can be determined based on the storage capacity and the acquired signal characteristics, and the embodiment of the present application does not make a unique limitation.

[0112] Furthermore, the acquired signal features are grouped into a feature sequence, which is input into the cylinder head vibration signal prediction model for prediction, and a signal feature prediction sequence is obtained, which includes at least one signal feature at a future time point. For the signal feature of each future time point in the signal feature prediction sequence, steps S130-170 are executed to determine whether there is a fault in the signal feature at each prediction time point. If there is a fault, the corresponding fault type and fault degree are provided to the staff to provide a cylinder head state prediction function, so as to achieve the effect of advance prediction and advance reminder, so that the staff can take the initiative to prevent the cylinder head and internal combustion engine in time before the fault occurs, and minimize the impact of the fault on the operation of the internal combustion engine.

[0113] In one possible implementation, the training process of the cylinder head vibration signal prediction model includes: obtaining at least one set of historical cylinder head vibration signals; extracting the signal features at each time point in the historical cylinder head vibration signals to obtain at least one historical signal feature sequence; dividing the historical signal feature sequence into a preceding feature sequence and a succeeding feature sequence, and the time point of each signal feature in the preceding feature sequence is earlier than the time point of each signal feature in the succeeding feature sequence.

[0114] Based on the preceding feature sequence and the succeeding feature sequence, the prediction model to be trained is trained to obtain a signal prediction model, wherein the training process uses the preceding feature sequence as a feature value and the succeeding feature sequence corresponding to the preceding feature sequence as a target value for training.

[0115] Call the signal prediction model to perform prediction processing on any preceding feature sequence to obtain a predicted feature sequence; determine whether the error between the subsequent feature sequence corresponding to the preceding feature sequence and the predicted feature sequence is greater than a preset error; if the error between the subsequent feature sequence and the predicted feature sequence is not greater than the preset error, determine that the signal prediction model is a trained cylinder head vibration signal prediction model; if the error between the subsequent feature sequence and the predicted feature sequence is greater than the preset error, adjust the model parameters of the signal prediction model to obtain a prediction model to be trained, and return to execute the steps of training the prediction model to be trained based on the preceding feature sequence and the subsequent feature sequence to obtain a signal prediction model, until the error between the subsequent feature sequence and the predicted feature sequence is not greater than the preset error.

[0116] Historical signal feature sequence , where F(t i ) represents the signal characteristics of the i-th historical cylinder head vibration signal, N t is the total number of historical cylinder head vibration signals. Divide the historical signal feature sequence into input sequence (predecessor feature sequence) and output sequence (post-order feature sequence), where 1 <m<N t . Get as many groups of historical signal feature sequences as possible, and divide the sequences according to the above method to provide training data for model training.

[0117] Furthermore, the input sequence As the input of the prediction model to be trained, the output sequence The output of the prediction model to be trained is used for training to obtain a signal prediction model.

[0118] In an embodiment of the present application, an LSTM model is used to construct a prediction model to be trained and trained, wherein the LSTM model includes an input layer, an LSTM layer, a fully connected layer, and an output layer.

[0119] Furthermore, the embodiment of the present application uses a teacherforcing method to update the model parameters of the signal prediction model to minimize the error between the prediction result and the corresponding output sequence, specifically including:

[0120] With any input sequence As the input of the signal prediction model, the prediction feature sequence is obtained , further, the error between the predicted feature sequence and the output sequence is calculated, wherein the error function defined in the embodiment of the present application is the following formula (5):

[0121] (5)

[0122] Among them, W is the model parameter of the LSTM model.

[0123] Use the back propagation algorithm and gradient descent method to minimize the error function, that is, min W L(W), repeat the above training process and update the model parameter W until the error between the predicted feature sequence and the output sequence output by the signal prediction model converges, or the number of training rounds reaches the preset number of rounds, then stop training the model and obtain a trained cylinder head vibration signal prediction model.

[0124] In another possible implementation, the process of extracting signal features from a cylinder head vibration signal may include: calling a filter to filter out noise within a preset frequency range in the vibration signal to obtain a first vibration signal; eliminating data points in the first vibration signal that meet preset elimination conditions to obtain a second vibration signal, wherein the preset elimination conditions include at least: an amplitude greater than a preset amplitude threshold, and a time interval between data points less than a preset interval threshold; separating the vibration signal of the cylinder head from the second vibration signal to obtain a cylinder head vibration signal to be detected.

[0125] The filter can be of different types according to the requirements, such as low-pass filter, bandpass filter, digital filter, etc., and the upper and lower limits of the filter can be set according to the frequency range of the historical cylinder head vibration signal to filter out high-frequency noise and low-frequency noise in the vibration signal. Optionally, in the embodiment of the present application, a sixth-order Butterworth bandpass filter is used to filter the vibration signal to obtain the first vibration signal x f (t), where the lower limit frequency is set to 1 kHz and the upper limit frequency is set to 10 kHz, as shown in the following equation (6), which is the transfer function of the sixth-order Butterworth bandpass filter:

[0126] (6)

[0127] Among them, f c is the cut-off frequency, and s is the complex frequency variable.

[0128] Furthermore, the abnormal data points in the first vibration signal are removed. It is understandable that the vibration signal is composed of a plurality of data points at different times and at the same time, and there will inevitably be factors such as unstable operation and abnormal vibration during the operation of the unit, resulting in the presence of data points in the collected vibration signal that deviate from the fluctuation range of the normal vibration signal. Therefore, in order to ensure the accuracy of the vibration signal, it is necessary to remove the abnormal data points in the collected vibration signal that deviate from the normal fluctuation range.

[0129] According to the normal fluctuation range of the cylinder head vibration signal, the screening conditions (or preset elimination conditions) for the data points are preset, such as the amplitude of the data point is greater than the preset amplitude threshold, and the time interval between the data points is less than the preset interval threshold, etc. The abnormal data points are screened out according to the preset screening conditions and eliminated, so as to obtain a second vibration signal with relatively normal fluctuation.

[0130] Optionally, the first vibration signal Setting the amplitude threshold and time interval threshold , then the preset elimination condition can be: and point , is determined as an abnormal point and removed to obtain the second vibration signal. and It can be set based on experience, such as .

[0131] It can be understood that the second vibration signal is a vibration signal with relatively normal fluctuations after noise is removed, but the vibration signal may also include vibration signals of other components, such as vibration signals of valves and camshafts adjacent to the cylinder head. Therefore, it is necessary to classify the signals in the second vibration signal according to the components or structures that generate the vibration signals, such as the vibration signals of the valves, the vibration signals of the camshafts, and the vibration signals of the cylinder head, and separate the vibration signal of the cylinder head from them to obtain the vibration signal of the cylinder head to be detected.

[0132] In one possible implementation, the process of separating the cylinder head vibration signal from the second vibration signal may include: segmenting the second vibration signal according to a preset segmentation duration to obtain multiple segmented signals; centering and whitening each segmented signal respectively to obtain a segmented signal to be analyzed corresponding to each segmented signal; calling an independent component analysis algorithm to analyze and process each segmented signal to be analyzed to obtain a separation matrix; based on the timing relationship of the time range corresponding to each segmented signal, determining the segmented signal with the earliest timing in the time range as the target segmented signal; and taking the product of the separation matrix and the target segmented signal as the cylinder head vibration signal to be detected.

[0133] First, the second vibration signal is segmented according to the preset segmentation time length T to obtain m segmented signals, x i (t), i=1,2,…,m.

[0134] For each segment signal x i (t) Perform centering and whitening to obtain the segmented signal z to be analyzed corresponding to each segmented signal i (t), centering can eliminate the bias of the segmented signal, and whitening can make each component of the segmented signal have the same variance. Therefore, zi The expectation (mean) of (t) is zero, z i The covariance matrix of (t) is equal to the unit matrix I, which means that the components of the second vibration signal are independent of each other and the variance of each component is normalized to 1.

[0135] Specifically, the FastICA blind source separation algorithm is called to analyze the segmented signal z i (t) is used as the input of the FastICA blind source separation algorithm. The goal of the FastICA blind source separation algorithm is to determine a separation matrix W such that In The components of i (t) is the result of whitening and centering. Therefore, the FastICA blind source separation algorithm can focus on maximizing the non-Gaussian type by minimizing mutual information or maximizing negative entropy without considering variance and covariance. Therefore, for the subsequent call of the FastICA blind source separation algorithm, the segmented signal z after centering and whitening is i (t) It can simplify the subsequent signal separation process and improve the computational efficiency of the FastICA blind source separation algorithm.

[0136] The FastICA blind source separation algorithm focuses on maximizing the non-Gaussian process, including: randomly initializing the separation matrix ,make .renew , so that The non-Gaussianity of is maximized, and the following formula (7) is the iterative formula for updating W:

[0137] (7)

[0138] in, , is the learning rate, and k represents the number of updates or iterations of W.

[0139] Repeatedly update W until the above formula (7) converges and obtains the optimal separation matrix W * The target separation signal corresponding to each segmented signal is obtained by the following formula (8): .

[0140] (8)

[0141] In the present application example, As the cylinder head vibration signal to be detected.

[0142] In a possible implementation, a corresponding fault warning method is determined according to the fault type and fault degree in the cylinder head diagnosis report; based on the fault warning method, information indicating the type of fault in the cylinder head of the internal combustion engine unit is sent to the front end.

[0143] In this example, the faults can be graded in advance according to their severity, such as minor faults, moderate faults, and major faults, and different fault warning methods can be set for each level. For example, for minor warnings, normal output methods such as emails, text messages, and display devices can be used, while for major warnings, telephone notifications can be used to remind staff so that they can respond to the fault in a timely manner for maintenance.

[0144] In another possible implementation, when the preliminary cylinder head diagnosis result is a label indicating that the cylinder head is faulty, a target repair plan corresponding to the target fault type and fault degree is matched from a pre-established fault repair plan repository and output.

[0145] The embodiment of the present application requires the staff to pre-store maintenance / repair plans corresponding to different fault types and different fault degrees in the processor in advance. Optionally, a large model with strong learning ability can be introduced. The large model matches the maintenance plan that can solve the current fault from the stored maintenance plans based on the diagnosis result of the current cylinder head, and outputs it to the staff to provide maintenance direction for the staff.

[0146] In summary, the cylinder head fault diagnosis method provided in the embodiment of the present application, based on the characteristics of the cylinder head vibration signal, separates interference signals such as noise, abnormal data, and other types of vibration signals from the collected vibration signal to obtain an accurate cylinder head vibration signal, and designs a signal feature extraction algorithm suitable for non-stationary, nonlinear complex signals. Based on the highly sensitive and robust cylinder head fault diagnosis model, fault diagnosis is performed on the signal features, which greatly improves the diagnostic accuracy of cylinder head faults.

[0147] The embodiment of the present application introduces deep learning theory into the field of cylinder head vibration signal analysis, and uses deep network structures such as convolutional neural networks (CNN) to achieve automatic learning of fault characteristics and end-to-end recognition of states, avoiding the complex manual feature engineering in traditional methods, greatly simplifying the signal processing process, and significantly improving the real-time and computational efficiency of fault diagnosis. The diagnosis time can be shortened to 1 / 5 of the original time.

[0148] The embodiment of the present application also proposes a future-oriented prediction method for cylinder head signal characteristics. By mining the trend change rules of the cylinder head vibration signal, a mapping model between the vibration characteristics at the current moment and the vibration characteristics at the future moment is established, thereby achieving early warning and prediction of cylinder head failures. Potential abnormalities of the cylinder head can be discovered at least 1-2 months in advance, transforming the maintenance plan from passive response to active prevention.

[0149] Optionally, the embodiment of the present application adopts a deep transfer learning strategy based on massive historical cylinder head vibration signal data. By mining the common characteristics of cylinder head vibration signals under different models and different operating conditions, a universal diagnostic model is constructed. It can also realize the transfer of diagnostic knowledge from one unit to another, greatly improving the scope of application of the diagnostic method, reducing the dependence on sample data, and providing the possibility for realizing status assessment between different units.

[0150] Based on this, a cylinder head fault diagnosis system developed in an embodiment of the present application realizes the full process automation from data acquisition, signal processing to fault assessment and maintenance decision-making. It can be deployed on-site at the unit to realize online diagnosis and real-time early warning of faults, effectively guide the optimization of maintenance strategies, reduce unplanned downtime, improve equipment availability, and extend the service life of the unit, with significant economic and social benefits.

[0151] The cylinder head fault diagnosis device provided in an embodiment of the present application is described below. The cylinder head fault diagnosis device described below and the cylinder head fault diagnosis method described above can be referenced to each other.

[0152] First, combine Figure 3 , a cylinder head fault diagnosis device for a processor used in a cylinder head fault diagnosis system is introduced, such as Figure 3 As shown, the cylinder head fault diagnosis device may include:

[0153] The signal acquisition unit 100 is used to acquire the vibration signal at the current time point collected by the vibration sensor during the operation of the internal combustion engine unit;

[0154] A signal elimination unit 200, used to eliminate interference signals in the vibration signal to obtain a cylinder head vibration signal to be detected;

[0155] A feature extraction unit 300, used to extract signal features from the cylinder head vibration signal to be detected;

[0156] The model processing unit 400 is used to call a pre-trained cylinder head fault diagnosis model, process the signal features, and obtain a preliminary cylinder head diagnosis result, wherein the cylinder head fault diagnosis model uses the signal feature samples of the cylinder head vibration signal samples as training samples, the training samples are correspondingly marked with sample labels, and the training is performed with the sample labels as output, and the sample labels are labels that characterize whether the signal feature samples have a fault;

[0157] a type determination unit 500, configured to, when the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault, perform similarity matching between the signal feature and a fault signal feature corresponding to each pre-stored fault type, obtain a target fault signal feature whose similarity with the signal feature reaches a preset threshold, and determine the fault type corresponding to the target fault signal feature as a target fault type;

[0158] A degree determination unit 600, for determining the degree of failure of the cylinder head according to the identified discrete degree of the cylinder head vibration signal to be detected;

[0159] The result output unit 700 is used to generate and output a cylinder head diagnosis report of the internal combustion engine unit based on the preliminary cylinder head diagnosis result, the target fault type and the fault degree.

[0160] In a possible implementation, the method further includes:

[0161] A signal feature acquisition unit, used to acquire signal features corresponding to a preset number of time points before the current time point;

[0162] A feature sequence generating unit, configured to generate a feature sequence of a signal to be predicted based on signal features corresponding to a preset number of time points before the current time point and the signal features corresponding to the current time point;

[0163] a feature sequence prediction unit, for calling a pre-trained cylinder head vibration signal prediction model, performing prediction processing on the feature sequence of the signal to be predicted, and obtaining a signal feature prediction sequence, wherein the signal feature prediction sequence includes: a signal feature of a cylinder head vibration signal at at least one prediction time point, and the cylinder head vibration signal prediction model is a model obtained by training with the signal feature sequence of the historical cylinder head vibration signal in the previous order as a feature value and the signal feature sequence of the historical cylinder head vibration signal in the subsequent order as a target value among the historical cylinder head vibration signals sorted according to the time series;

[0164] The fault prediction unit is used to call the pre-trained cylinder head fault diagnosis model, process the signal features corresponding to each prediction time point in the signal feature prediction sequence, and obtain a preliminary prediction diagnosis result corresponding to each prediction time point.

[0165] In one possible implementation, the unit for training the cylinder head vibration signal prediction model includes:

[0166] A historical signal acquisition unit, used to acquire at least one set of historical cylinder head vibration signals;

[0167] A feature sequence extraction unit, used to extract the signal feature of each time point in the historical cylinder head vibration signal to obtain at least one historical signal feature sequence;

[0168] A sequence division unit, used for dividing the historical signal feature sequence into a pre-sequence feature sequence and a post-sequence feature sequence, wherein the time point of each signal feature in the pre-sequence feature sequence is earlier than the time point of each signal feature in the post-sequence feature sequence;

[0169] A prediction model training unit, used for training the prediction model to be trained based on the preceding feature sequence and the succeeding feature sequence to obtain a signal prediction model, wherein the training process uses the preceding feature sequence as a feature value and uses the succeeding feature sequence corresponding to the preceding feature sequence as a target value for training;

[0170] A sequence prediction unit, used for calling the signal prediction model to perform prediction processing on any of the preceding feature sequences to obtain a predicted feature sequence;

[0171] An error judgment unit, used to judge whether the error between the subsequent feature sequence corresponding to the preceding feature sequence and the predicted feature sequence is greater than a preset error;

[0172] a model determination unit, configured to determine that the signal prediction model is a trained cylinder head vibration signal prediction model if the judgment result of the error judgment unit is no;

[0173] A jump training unit is used to adjust the model parameters of the signal prediction model if the judgment result of the error judgment unit is yes, to obtain the prediction model to be trained, and return to the prediction model training unit to execute the step of training the prediction model to be trained based on the preceding feature sequence and the succeeding feature sequence to obtain the signal prediction model, until the error between the succeeding feature sequence and the prediction feature sequence is no greater than the preset error.

[0174] In a possible implementation, the unit for training the cylinder head fault diagnosis model includes:

[0175] A sample signal acquisition unit, used to acquire at least one set of cylinder head vibration signal samples and at least one fault signal feature corresponding to a fault type;

[0176] A sample feature extraction unit, used to extract the signal feature of each time point in the cylinder head vibration signal sample to obtain a signal feature set;

[0177] A labeling unit, configured to determine the sample label corresponding to each signal feature in the feature set based on the fault signal feature corresponding to the fault type, wherein the sample label includes: a label indicating whether a fault exists;

[0178] A model training unit is used to train the fault diagnosis model to be trained based on the feature set and the sample label corresponding to each signal feature in the feature set to obtain a cylinder head fault diagnosis model, wherein the training process includes: using the signal feature sample of the cylinder head vibration signal sample as the feature value and using the sample label corresponding to the signal feature sample as the target value for training.

[0179] In a possible implementation, the signal removal unit 200 includes:

[0180] A filter subunit, used to call a filter to filter out noise within a preset frequency range in the vibration signal to obtain a first vibration signal;

[0181] A data point elimination subunit is used to eliminate data points that meet preset elimination conditions in the first vibration signal to obtain a second vibration signal, wherein the preset elimination conditions at least include: an amplitude greater than a preset amplitude threshold, and a time interval between data points less than a preset interval threshold;

[0182] The signal separation subunit is used to separate the vibration signal of the cylinder head from the second vibration signal to obtain the cylinder head vibration signal to be detected.

[0183] In a possible implementation, the signal separation subunit includes:

[0184] A signal segmentation subunit, configured to segment the second vibration signal according to a preset segmentation duration to obtain a plurality of segmented signals;

[0185] The segmented signal processing subunit is used to respectively center and whiten each of the segmented signals to obtain a segmented signal to be analyzed corresponding to each of the segmented signals;

[0186] The algorithm calling subunit is used to call the independent component analysis algorithm to analyze and process each of the segmented signals to be analyzed to obtain a separation matrix;

[0187] A target segment signal determination subunit, configured to determine the segment signal that is the earliest in the time sequence of the time range as the target segment signal according to the time sequence relationship of the time ranges corresponding to each of the segment signals;

[0188] The cylinder head signal determination subunit is used to take the product of the separation matrix and the target segmented signal as the cylinder head vibration signal to be detected.

[0189] In a possible implementation, the method further includes:

[0190] a warning mode determination unit, configured to determine a corresponding fault warning mode according to the fault type and the fault degree described in the cylinder head diagnosis report;

[0191] The early warning output unit is used to send information indicating that a fault of the fault type exists in the cylinder head of the internal combustion engine unit to the front end based on the fault early warning method.

[0192] In a possible implementation, the method further includes:

[0193] The maintenance plan output unit is used to match and output a target maintenance plan corresponding to the target fault type and the fault degree from a pre-established fault maintenance plan repository when the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault.

[0194] Based on this, a cylinder head fault diagnosis device provided in an embodiment of the present application can realize the automation of the entire process from data collection, signal processing to fault assessment and maintenance decision-making, and remove interference signals from the vibration signals collected by the vibration sensor to obtain the cylinder head vibration signal used for subsequent fault diagnosis, thereby avoiding vibration interference with adjacent components of the cylinder head, such as valves, camshafts, etc. Furthermore, the pre-trained cylinder head fault diagnosis model is used to perform quantitative fault state diagnosis on the signal characteristics of the cylinder head vibration signal, which can improve the objectivity of fault diagnosis and the accuracy and precision of cylinder head fault diagnosis based on the cylinder head vibration signal compared to human subjective judgment.

[0195] In addition, when the preliminary diagnostic results indicate that there is a fault in the cylinder head, the signal characteristics of the cylinder head are further matched with the fault signal characteristics of each fault type for similarity and the degree of discreteness is identified to achieve quantitative verification of the fault type and degree of the cylinder head, avoiding the complex manual feature analysis process in the traditional method, greatly simplifying the signal processing process, and significantly improving the diagnostic efficiency.

[0196] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0197] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0198] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cylinder head fault diagnosis method, characterized in that: A processor applied to a cylinder head fault diagnosis system, the cylinder head fault diagnosis system at least further comprising: a vibration sensor provided on the cylinder head when collecting vibration signals, the cylinder head fault diagnosis method comprising: During the operation of the internal combustion engine unit, obtaining a vibration signal at a current time point collected by the vibration sensor; Eliminating interference signals in the vibration signal to obtain a cylinder head vibration signal to be detected; Extracting signal features from the cylinder head vibration signal to be detected; Calling a pre-trained cylinder head fault diagnosis model, processing the signal features, and obtaining a preliminary cylinder head diagnosis result, wherein the cylinder head fault diagnosis model uses signal feature samples of cylinder head vibration signal samples as training samples, the training samples are correspondingly marked with sample labels, and the training is performed using the sample labels as output, wherein the sample labels are labels that characterize whether the signal feature samples have a fault; In the case where the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault, the signal feature is matched with a pre-stored fault signal feature corresponding to each fault type to obtain a target fault signal feature whose similarity with the signal feature reaches a preset threshold, and the fault type corresponding to the target fault signal feature is determined as the target fault type; Determining the degree of failure of the cylinder head according to the identified discrete degree of the cylinder head vibration signal to be detected; A cylinder head diagnosis report of the internal combustion engine group is generated and outputted based on the preliminary cylinder head diagnosis result, the target fault type and the fault degree.

2. The cylinder head fault diagnosis method according to claim 1, characterized in that: After extracting the signal feature from the cylinder head vibration signal to be detected, the method further includes: Obtaining signal characteristics corresponding to a preset number of time points before the current time point; Generate a signal feature sequence to be predicted based on signal features corresponding to a preset number of time points before the current time point and the signal features corresponding to the current time point; Calling a pre-trained cylinder head vibration signal prediction model, performing prediction processing on the signal feature sequence to be predicted, and obtaining a signal feature prediction sequence, wherein the signal feature prediction sequence includes: a signal feature of a cylinder head vibration signal at at least one prediction time point, and the cylinder head vibration signal prediction model is a model obtained by training with the signal feature sequence of the historical cylinder head vibration signals in the previous order as feature values ​​and the signal feature sequence of the historical cylinder head vibration signals in the subsequent order as target values ​​in the historical cylinder head vibration signals sorted according to the time series; The pre-trained cylinder head fault diagnosis model is called to process the signal features corresponding to each prediction time point in the signal feature prediction sequence respectively to obtain a preliminary prediction diagnosis result corresponding to each prediction time point.

3. The cylinder head fault diagnosis method according to claim 2, characterized in that: The training process of the cylinder head vibration signal prediction model includes: Acquire at least one set of historical cylinder head vibration signals; Extracting the signal feature at each time point in the historical cylinder head vibration signal to obtain at least one historical signal feature sequence; Dividing the historical signal feature sequence into a preceding feature sequence and a succeeding feature sequence, wherein the time point of each signal feature in the preceding feature sequence is earlier than the time point of each signal feature in the succeeding feature sequence; Based on the preceding feature sequence and the succeeding feature sequence, a prediction model to be trained is trained to obtain a signal prediction model, wherein the training process uses the preceding feature sequence as a feature value and uses the succeeding feature sequence corresponding to the preceding feature sequence as a target value for training; Calling the signal prediction model to perform prediction processing on any of the preceding feature sequences to obtain a predicted feature sequence; Determine whether the error between the subsequent feature sequence corresponding to the preceding feature sequence and the predicted feature sequence is greater than a preset error; If the error between the subsequent feature sequence and the predicted feature sequence is not greater than the preset error, determining that the signal prediction model is a trained cylinder head vibration signal prediction model; If the error between the subsequent feature sequence and the predicted feature sequence is greater than the preset error, the model parameters of the signal prediction model are adjusted to obtain the prediction model to be trained, and the step of training the prediction model to be trained based on the preceding feature sequence and the subsequent feature sequence to obtain the signal prediction model is returned to execute until the error between the subsequent feature sequence and the predicted feature sequence is no greater than the preset error.

4. The cylinder head fault diagnosis method according to claim 1, characterized in that: The training process of the cylinder head fault diagnosis model includes: Acquire at least one set of cylinder head vibration signal samples and at least one fault signal feature corresponding to a fault type; Extracting the signal feature at each time point in the cylinder head vibration signal sample to obtain a signal feature set; Based on the fault signal feature corresponding to the fault type, determining the sample label corresponding to each signal feature in the feature set, the sample label including: a label indicating whether there is a fault; Based on the feature set and the sample labels corresponding to each signal feature in the feature set, the fault diagnosis model to be trained is trained to obtain a cylinder head fault diagnosis model, wherein the training process includes: using the signal feature samples of the cylinder head vibration signal samples as feature values, and using the sample labels corresponding to the signal feature samples as target values ​​for training.

5. The cylinder head fault diagnosis method according to claim 1, characterized in that: The step of removing the interference signal from the vibration signal to obtain the cylinder head vibration signal to be detected includes: Calling a filter to filter out noise within a preset frequency range in the vibration signal to obtain a first vibration signal; Eliminate data points that meet preset elimination conditions in the first vibration signal to obtain a second vibration signal, wherein the preset elimination conditions at least include: an amplitude greater than a preset amplitude threshold, and a time interval between data points less than a preset interval threshold; The vibration signal of the cylinder head is separated from the second vibration signal to obtain the cylinder head vibration signal to be detected.

6. The cylinder head fault diagnosis method according to claim 5, characterized in that: The step of separating the vibration signal of the cylinder head from the second vibration signal to obtain the vibration signal of the cylinder head to be detected comprises: Segmenting the second vibration signal according to a preset segment duration to obtain a plurality of segment signals; Centering and whitening each of the segmented signals respectively to obtain a segmented signal to be analyzed corresponding to each of the segmented signals; Invoke an independent component analysis algorithm to analyze and process each of the segmented signals to be analyzed to obtain a separation matrix; According to the timing relationship of the time ranges corresponding to each of the segmented signals, determining the segmented signal with the earliest timing in the time range as the target segmented signal; The product of the separation matrix and the target segmented signal is used as the cylinder head vibration signal to be detected.

7. The cylinder head fault diagnosis method according to claim 1, characterized in that: Also includes: Determining a corresponding fault warning method according to the fault type and the fault degree described in the cylinder head diagnostic report; Based on the fault warning method, information indicating that a fault of the fault type exists in the cylinder head of the internal combustion engine unit is sent to the front end.

8. The cylinder head fault diagnosis method according to claim 7, characterized in that: Also includes: When the preliminary cylinder head diagnosis result is a label indicating that the cylinder head is faulty, a target repair solution corresponding to the target fault type and the fault degree is matched from a pre-established fault repair solution repository and outputted.

9. A cylinder head fault diagnosis device, characterized in that: A processor applied to a cylinder head fault diagnosis system, the cylinder head fault diagnosis system at least further comprising: a vibration sensor provided on the cylinder head when collecting vibration signals, the cylinder head fault diagnosis device comprising: A signal acquisition unit, used to acquire the vibration signal at a current time point collected by the vibration sensor during the operation of the internal combustion engine unit; A signal elimination unit, used for eliminating interference signals in the vibration signal to obtain a cylinder head vibration signal to be detected; A feature extraction unit, used to extract signal features from the cylinder head vibration signal to be detected; a model processing unit, used to call a pre-trained cylinder head fault diagnosis model, process the signal features, and obtain a preliminary cylinder head diagnosis result, wherein the cylinder head fault diagnosis model uses a signal feature sample of a cylinder head vibration signal sample as a training sample, the training sample is correspondingly marked with a sample label, and the training is performed using the sample label as an output, wherein the sample label is a label that characterizes whether the signal feature sample has a fault; a type determination unit, configured to, when the preliminary cylinder head diagnosis result is a label indicating that the cylinder head has a fault, perform similarity matching between the signal feature and a fault signal feature corresponding to each pre-stored fault type, obtain a target fault signal feature whose similarity with the signal feature reaches a preset threshold, and determine the fault type corresponding to the target fault signal feature as a target fault type; a degree determination unit, configured to determine the degree of failure of the cylinder head according to the identified discrete degree of the vibration signal of the cylinder head to be detected; The result output unit is used to generate and output a cylinder head diagnosis report of the internal combustion engine unit based on the preliminary cylinder head diagnosis result, the target fault type and the fault degree.

10. A cylinder head fault diagnosis system, characterized in that: include: When collecting vibration signals, a vibration sensor is provided on the cylinder head; A processor, used to implement the cylinder head fault diagnosis method described in any one of claims 1-8.

Images