Methods, devices and equipment for engine misfire fault identification under data security collaboration
By initializing and aggregating the engine misfire fault identification model on the client side, pre-training and classifying it using local high-quality vibration signals, and aggregating model parameters by combining aggregation influencing factors, the problems of misdiagnosis and robustness in engine misfire fault detection in traditional methods are solved, thereby improving the identification accuracy and model performance.
Patent Information
- Application Number
- CN202410575952.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-10
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2044-05-10
AI Technical Summary
In the existing technology, the detection methods for engine misfire faults have the problems of misdiagnosis or missed diagnosis, and the robustness is poor. In particular, the accuracy of multi-cylinder misfire fault diagnosis is insufficient under high speed and low load conditions. In addition, traditional methods require a lot of test resources and have a long calibration and development cycle.
By initializing the engine misfire fault identification model on multiple clients, pre-training the model using local high-quality vibration signals, classifying and aggregating the pre-trained model parameters, determining the aggregation category, and combining the aggregation influence factor to aggregate and iteratively update the model parameters, a target engine misfire fault identification model is formed.
While ensuring data privacy and security, the accuracy of the engine misfire fault identification model has been improved. It can adjust the strategy in real time according to the actual situation, enhance data quality and model performance, and promote knowledge sharing and cooperation among clients.
Smart Images

Figure CN118520253B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of engine technology, and in particular to a method, apparatus and equipment for identifying engine misfire faults under data security coordination. Background Technology
[0002] Engine misfire refers to the phenomenon where the air-fuel mixture in the cylinder fails to burn or burns poorly due to hardware malfunctions in the intake and exhaust systems, fuel supply systems, ignition systems, or other reasons. Mild engine misfires can worsen emissions or even exceed standards, while severe misfires can lead to increased fuel consumption, reduced power output, and affect engine smoothness, power, and fuel economy, and may even damage the three-way catalytic converter. Therefore, engine misfire diagnosis is an important component of on-board diagnostic systems.
[0003] Currently, the commonly used method for detecting engine misfire faults is the crankshaft instantaneous speed fluctuation method. This method has advantages such as convenient measurement and low cost, and has been widely studied and applied abroad. However, acceleration and deceleration, load changes, and other factors can also cause abnormal crankshaft speed fluctuations. Therefore, the causes of abnormal speed fluctuations are not only due to misfire faults, but also to factors such as acceleration and deceleration, load changes, etc. At the same time, this method has poor accuracy in diagnosing misfire faults and multi-cylinder misfire faults under high speed and low load conditions. On the other hand, in the process of developing engine misfire diagnosis systems, on-board misfire diagnosis systems need to rely on a large number of bench tests and whole vehicle road tests, and set different misfire modes to complete misfire diagnosis calibration under different operating conditions. However, this method requires a large number of test resources, has a long calibration development cycle, and is inefficient. At the same time, OBD calibration engineers need to set misfire thresholds based on experience, but the calibrated misfire thresholds often cannot cover all actual road conditions, which may lead to failures such as misdiagnosis or missed diagnosis, resulting in poor robustness of misfire diagnosis.
[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention
[0005] The main objective of this application is to provide a method, apparatus, and device for identifying engine misfire faults under data security collaboration, aiming to solve the technical problem that traditional methods in the prior art are unable to accurately determine the engine misfire fault mode.
[0006] To achieve the above objectives, this application provides a method for identifying engine misfire faults under data security coordination, comprising:
[0007] The initial model parameters are distributed to multiple clients so that the clients can initialize the engine misfire fault identification model based on the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals.
[0008] Obtain the pre-trained model parameters of the engine misfire fault identification model uploaded by the client, classify the engine misfire fault identification model based on the pre-trained model parameters, and determine the aggregation category of the engine misfire fault identification model in the client.
[0009] Based on local high-quality vibration signals, the engine misfire fault identification model is further trained. The training model parameters of the engine misfire fault identification model uploaded by the client are obtained. Based on the training model parameters and the aggregation category, the aggregation influence factor of the client is determined.
[0010] Based on the client's aggregation influence factor, the training model parameters of clients in the same aggregation category are aggregated to obtain the aggregation model parameters corresponding to each aggregation category.
[0011] The aggregated model parameters are sent to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model. Based on the target engine misfire fault identification model, misfire fault monitoring of the engine is performed.
[0012] In one embodiment, the step of classifying the engine misfire fault identification model based on pre-trained model parameters and determining the aggregation category of the engine misfire fault identification model includes:
[0013] Based on the sample distance between pre-trained model parameters, determine the baseline distance and sample density of the pre-trained model parameters;
[0014] Based on the baseline distance and sample density of the pre-trained model parameters, the pre-trained model parameters are divided into corresponding sample clusters;
[0015] Based on the sample clusters of pre-trained model parameters, the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters is determined.
[0016] In one embodiment, the step of determining the baseline distance and sample tightness of the pre-trained model parameters based on the sample distance between the pre-trained model parameters includes:
[0017] Obtain the first correspondence between the sample distance, the baseline distance coefficient, the number of pre-trained model parameters, and the baseline distance;
[0018] The baseline distance is determined based on the sample distance between pre-trained model parameters, the baseline distance coefficient, the number of pre-trained model parameters, and the first correspondence.
[0019] Based on the sample distance and the baseline distance, determine the close parameters corresponding to the pre-trained model parameters. The sample distance between the pre-trained model parameters and the corresponding close parameters is less than the baseline distance.
[0020] To obtain a second correspondence between the number of tight parameters, the number of pre-trained model parameters, and sample tightness;
[0021] The sample density of the pre-trained model parameters is obtained based on the number of closely corresponding parameters of the pre-trained model parameters, the number of pre-trained model parameters, and the second correspondence.
[0022] In one embodiment, the step of assigning pre-trained model parameters to corresponding sample clusters based on the baseline distance and sample density of the pre-trained model parameters includes:
[0023] Based on the sample density of the pre-trained model parameters, the pre-trained model parameters are sorted in descending order to obtain a parameter sequence. The first pre-trained model parameter in the parameter sequence is added to the center point set as the initial cluster center point.
[0024] The center point screening threshold is determined based on the cluster center difference parameter and the baseline distance;
[0025] Obtain the center distance between the pre-trained model parameters and the initial cluster center points in the center point set. Select the pre-trained model parameters whose center distance is greater than the center point screening threshold as the initial cluster center points and add them to the center point set.
[0026] When the number of initial cluster center points in the center point set reaches a preset number, the cluster center point corresponding to the pre-trained model parameters is determined based on the center distance between the pre-trained model parameters and the initial cluster center points. The pre-trained model parameters are then assigned to the sample clusters of the corresponding cluster center points, and the distance between the pre-trained model parameters and the cluster center points in the sample clusters is determined.
[0027] Based on the distance between the pre-trained model parameters in the sample cluster and the cluster center point, the mean model parameters of the sample cluster are determined, and the mean model parameters of the sample cluster are used as the new cluster center point.
[0028] When the clustering conditions are met, the sample clusters corresponding to the pre-trained model parameters are determined based on the distance between the pre-trained model parameters and the cluster center points. Then, the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters is determined based on the sample clusters of the pre-trained model parameters.
[0029] If the clustering conditions are not met, return to the step of determining the distance between the pre-trained model parameters and the cluster centroid in the sample cluster.
[0030] In one embodiment, the step of determining the client's aggregation influence factor based on the trained model parameters and aggregation category includes:
[0031] Based on the aggregation category of the engine misfire fault identification model in the client, the cluster center point of the corresponding sample cluster in the client is determined, and the aggregation distance between the training model parameters of the client and the cluster center point is determined.
[0032] To obtain the third correspondence between aggregation distance and aggregation influence factor;
[0033] The aggregation influence factor of the client is determined based on the aggregation distance corresponding to the client and the third correspondence.
[0034] In one embodiment, the step of aggregating the training model parameters of clients within the same aggregation category based on the client's aggregation influence factor to obtain the aggregation model parameters corresponding to each aggregation category includes:
[0035] Obtain the fourth correspondence between the aggregated impact factor, training model parameters, number of signals from clients, total number of signals, and aggregated model parameters;
[0036] Based on the client's aggregation influence factor, the training model parameters of the engine misfire fault identification model, the number of signals from the client, the total number of signals, and the fourth correspondence, the aggregation model parameters corresponding to each aggregation category are determined.
[0037] In one embodiment, the step of sending the aggregated model parameters to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model includes:
[0038] Based on the aggregation category corresponding to the client, the aggregation model parameters are sent to the client so that the client can update the engine misfire fault identification model and determine whether the iteration termination condition is met.
[0039] When the iteration termination condition is met, the updated engine misfire fault identification model is used as the target engine misfire fault identification model, and the steps of monitoring engine misfire fault based on the target engine misfire fault identification model are executed.
[0040] If the iteration termination condition is not met, the engine misfire fault identification model is retrained based on the local high-quality vibration signal, and the training model parameters are updated. Then, the process returns to retrieve the training model parameters of the engine misfire fault identification model uploaded by the client, and determines the client's aggregation influence factor based on the training model parameters and aggregation category.
[0041] In one embodiment, the client's local high-quality vibration signal is provided with a corresponding training label. The training label is set according to the misfire fault mode of the local high-quality vibration signal. The misfire fault mode includes at least single-cylinder single misfire, single-cylinder continuous misfire, multi-cylinder single misfire, and multi-cylinder continuous misfire.
[0042] In one embodiment, different misfire failure modes correspond to different optimization strategies, which are used to optimize the engine.
[0043] Furthermore, to achieve the above objectives, this application also proposes an engine misfire fault identification device under data security collaboration, which includes:
[0044] The model training module is used to distribute the initial model parameters to multiple clients so that the clients can initialize the engine misfire fault identification model according to the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals.
[0045] The model pre-clustering module is used to obtain the pre-trained model parameters of the engine misfire fault identification model uploaded by the client, classify the engine misfire fault identification model based on the pre-trained model parameters, and determine the aggregation category of the engine misfire fault identification model in the client.
[0046] The aggregation factor calculation module is used to continue training the engine misfire fault identification model based on local high-quality vibration signals, obtain the training model parameters of the engine misfire fault identification model uploaded by the client, and determine the aggregation influence factor of the client based on the training model parameters and aggregation category.
[0047] The parameter aggregation module is used to aggregate the training model parameters of clients in the same aggregation category based on the client's aggregation influence factor, so as to obtain the aggregated model parameters corresponding to each aggregation category.
[0048] The misfire fault identification module is used to send the aggregated model parameters to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model. Based on the target engine misfire fault identification model, the engine misfire fault is monitored.
[0049] Furthermore, to achieve the above objectives, this application also proposes an engine misfire fault identification device under data security integration. The engine misfire fault identification device under data security integration includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the engine misfire fault identification method under data security integration as described above.
[0050] In addition, to achieve the above objectives, the present invention also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the engine misfire fault identification method under data security coordination as described above.
[0051] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the engine misfire fault identification method under data security coordination as described above.
[0052] This application provides a data-secure collaborative method for engine misfire fault identification. Initial model parameters are distributed to multiple clients, enabling each client to initialize an engine misfire fault identification model based on these parameters. The model is then pre-trained using local high-quality vibration signals. Pre-trained model parameters from client-uploaded models are obtained, and these parameters are used to classify the models and determine their aggregation categories. The model is further trained using local high-quality vibration signals, and training model parameters from client-uploaded models are obtained. Aggregation influence factors for each client are determined based on these parameters and the aggregation categories. The training model parameters from clients within the same aggregation category are aggregated using these aggregation influence factors to obtain aggregated model parameters for each category. These aggregated model parameters are then distributed to the corresponding clients, allowing them to iteratively update their engine misfire fault identification models to obtain a target model. Finally, engine misfire fault monitoring is performed based on this target model. Each client in this application utilizes high-quality local data to construct a local model. After aggregating the model parameters of each client, the resulting aggregated model parameters are distributed to each client for iterative model updates to obtain the final engine misfire fault identification model. During this process, by sharing model parameters from different clients—that is, sharing high-quality data while ensuring data privacy and security—data quality is enhanced, thereby improving the identification accuracy of the engine misfire fault identification model. Furthermore, the aggregation strategy can be changed in real time according to actual conditions, effectively filtering and utilizing clients beneficial to global model updates, further improving model performance and identification accuracy. In addition, classifying all participating clients helps participants discover similarities and correlations between data, promoting knowledge sharing and cooperation among participants, further improving the model's identification accuracy. This solves the technical problem of traditional methods being unable to accurately determine engine misfire fault modes, thus improving the identification accuracy of the engine misfire fault identification model. Attached Figure Description
[0053] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0055] Figure 1 This is a flowchart illustrating an embodiment of the engine misfire fault identification method under the data security collaboration of this application.
[0056] Figure 2 A simplified architectural diagram of the engine misfire fault identification method under data security coordination provided in Embodiment 1 of this application;
[0057] Figure 3 A simplified flowchart illustrating the engine misfire fault identification method under data security integration provided in Embodiment 1 of this application;
[0058] Figure 4 This is a schematic diagram of the module structure of the engine misfire fault identification device under data security integration in an embodiment of this application;
[0059] Figure 5 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the engine misfire fault identification method under data security coordination in the embodiments of this application.
[0060] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0061] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.
[0062] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.
[0063] The main solution of this application embodiment is as follows: Initial model parameters are distributed to multiple clients, enabling the clients to initialize an engine misfire fault identification model based on the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals; Pre-trained model parameters of the engine misfire fault identification model uploaded by the clients are obtained; based on the pre-trained model parameters, the engine misfire fault identification model is classified to determine the aggregation category of the engine misfire fault identification models in the clients; The engine misfire fault identification model is further trained based on local high-quality vibration signals, and the training model parameters of the engine misfire fault identification model uploaded by the clients are obtained; Based on the training model parameters and the aggregation category, the aggregation influence factor of the clients is determined; Based on the aggregation influence factor of the clients, the training model parameters of clients in the same aggregation category are aggregated to obtain the aggregation model parameters corresponding to each aggregation category; The aggregation model parameters are distributed to the corresponding clients, enabling the clients to iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model; Based on the target engine misfire fault identification model, engine misfire fault monitoring is performed.
[0064] Currently, the commonly used method for detecting engine misfire faults is the crankshaft instantaneous speed fluctuation method. This method has advantages such as convenient measurement and low cost, and has been widely studied and applied abroad. However, acceleration and deceleration, load changes, and other factors can also cause abnormal crankshaft speed fluctuations. Furthermore, abnormal speed fluctuations are not solely caused by misfire faults, but also by acceleration and deceleration, load changes, etc. Therefore, the diagnostic accuracy for misfires and multi-cylinder misfires under high-speed, low-load conditions is poor. On the other hand, the development of on-board misfire diagnostic systems requires extensive bench testing and vehicle road testing during engine misfire calibration, setting different misfire modes and completing misfire diagnostic calibration under various operating conditions. However, this method requires significant testing resources, has a long calibration development cycle, and is inefficient. Additionally, OBD calibration engineers need to set misfire thresholds based on experience, but these thresholds often cannot cover all actual road conditions, potentially leading to misdiagnosis or missed diagnosis, resulting in poor robustness of misfire diagnosis.
[0065] This application provides a solution in which each client utilizes high-quality local data to build a local model. After aggregating the model parameters of each client, the resulting aggregated model parameters are distributed to each client for iterative model updates to obtain the final engine misfire fault identification model. During this process, by sharing model parameters from different clients—that is, sharing high-quality data while ensuring data privacy and security—data quality is enhanced, thereby improving the identification accuracy of the engine misfire fault identification model. Furthermore, the aggregation strategy can be changed in real time according to actual conditions, effectively filtering and utilizing clients beneficial to global model updates, further improving model performance and identification accuracy. In addition, classifying all participating clients helps participants discover similarities and correlations between data, promoting knowledge sharing and cooperation among participants, further improving the model's identification accuracy. This solves the technical problem of traditional methods being unable to accurately determine engine misfire fault modes and improves the identification accuracy of the engine misfire fault identification model.
[0066] It should be noted that the execution subject of this embodiment is an engine misfire fault identification device under data security collaboration, which is a computing service device with data processing, network communication and program execution functions. The engine misfire fault identification device under data security collaboration usually needs to be a server that all clients can trust, such as a central server. This embodiment does not specifically limit this. The following uses a central server as an example to describe this embodiment and the following embodiments.
[0067] This application provides a method for identifying engine misfire faults under data security collaboration, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the engine misfire fault identification method under the data security collaboration of this application.
[0068] In this embodiment, the engine misfire fault identification method under data security collaboration includes steps S10 to S50:
[0069] Step S10: The initial model parameters are sent to multiple clients so that the clients initialize the engine misfire fault identification model according to the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals.
[0070] It should be noted that the client can be any factory / enterprise that needs to build an engine misfire fault identification model. In this embodiment, there are typically multiple clients, and the specific number can be determined according to the actual situation, without any specific limitation. The engine misfire fault identification model is a model used to identify engine misfire faults, and each client builds an engine misfire fault identification model. The initial model parameters are the initialization model parameters determined by the central server according to actual needs. The client uses the initial model parameters to initialize the engine misfire fault identification model. In this embodiment, a neural network model can be used as the base model for the engine misfire fault identification model, without any specific limitation.
[0071] Additionally, it should be noted that the engine in this embodiment is a gasoline engine, and the local high-quality vibration signal refers to the high-quality engine vibration signal collected by each client for training the model. During the engine's operating cycle, the instantaneous angular velocity of the crankshaft can effectively reflect the crankshaft's operating state and stability, the smoothness of engine operation, the uniformity of combustion in each cylinder, and can characterize the engine's torque. Therefore, this embodiment uses a crankshaft position sensor to obtain the instantaneous angular velocity of the crankshaft as the engine vibration signal for identification. After screening, high-quality vibration signals are selected as the local high-quality vibration signal.
[0072] Understandably, all vibration signals are collected under different misfire modes. Therefore, each engine vibration signal has a corresponding misfire fault mode. These misfire fault modes include at least single-cylinder misfire (single misfire of cylinder i), continuous single-cylinder misfire (continuous misfire of cylinder i), single-cylinder misfire (single misfire of cylinders i and j), and continuous multi-cylinder misfire (continuous misfire of cylinders i and j). Specifically, single-cylinder misfire refers to a misfire occurring once in cylinder i within one working cycle, followed by no further misfires during ignition in the subsequent five working cycles; this is considered an isolated incident. Continuous single-cylinder misfire refers to misfires occurring in cylinder i in two adjacent working cycles. Single-cylinder misfire refers to a misfire occurring once in cylinder i and once in cylinder j within one working cycle, followed by no further misfires during ignition in the subsequent five working cycles; this is considered an isolated incident. Continuous multi-cylinder misfire refers to misfires occurring in cylinders i and j in two adjacent working cycles. When training the model, all local high-quality vibration signals are labeled according to the fire fault modes corresponding to the local high-quality vibration signals. In other words, training labels are set according to the fire fault modes of the local high-quality vibration signals, and these training labels can reflect the fire fault modes of the local high-quality vibration signals.
[0073] It should be understood that the client obtains engine vibration signals of different misfire failure modes from the local machine, selects high-quality signals as the client's local high-quality vibration signals, and after the central server sends the initial model parameters to each client, the client builds the initial engine misfire failure identification model according to the initial model parameters. Then, the local high-quality vibration signals and the misfire failure mode labels of the local high-quality vibration signals are used as training data to pre-train the model.
[0074] Step S20: Obtain the pre-trained model parameters of the engine misfire fault identification model uploaded by the client; classify the engine misfire fault identification model based on the pre-trained model parameters; and determine the aggregation category of the engine misfire fault identification model in the client.
[0075] It should be noted that the pre-trained model parameters refer to the model parameters of the engine misfire fault identification model pre-trained in the client, which are uploaded by the client to the central server. The aggregation category is the category to which the engine misfire fault identification model belongs. For example, according to the number of cylinders, it can be divided into three-cylinder, four-cylinder, and six-cylinder. In this embodiment, the pre-trained model parameters are used to classify the engine misfire fault identification models in all clients to determine the category of each model.
[0076] In one feasible implementation, the step of classifying the engine misfire fault identification model based on pre-trained model parameters and determining the aggregation category of the engine misfire fault identification model may include steps S201 to S203:
[0077] Step S201: Based on the sample distance between the pre-trained model parameters, determine the baseline distance and sample density of the pre-trained model parameters;
[0078] It should be noted that the sample distance between pre-trained model parameters refers to the distance between any two pre-trained model parameters. In this case, the pre-trained model parameters can be considered as sample points. The formula for calculating the sample distance is as follows:
[0079]
[0080] In the formula, This represents the pre-trained model parameters of client i. For the pre-trained model parameters of client j, For pre-trained model parameters With W t j The sample distance between them.
[0081] In one feasible implementation, step S201 may include steps S2011 to S2015:
[0082] Step S2011: Obtain the first correspondence between the sample distance, the baseline distance coefficient, the number of pre-trained model parameters, and the baseline distance;
[0083] It should be noted that the baseline distance is the distance set as the benchmark in this embodiment, and the baseline distance coefficient is the correlation coefficient that needs to be considered when calculating the baseline distance. The default value is 1, and it can be flexibly adjusted according to the actual situation. The first correspondence between the sample distance, the baseline distance coefficient, the number of pre-trained model parameters, and the baseline distance refers to the calculation formula of the baseline distance, as shown below:
[0084]
[0085] In the formula, L B The reference distance is represented by ε, which represents the reference distance coefficient. For pre-trained model parameters With W t j The sample distance between them, where n represents the number of parameters in the pre-trained model.
[0086] Step S2012: Determine the baseline distance based on the sample distance between pre-trained model parameters, the baseline distance coefficient, the number of pre-trained model parameters, and the first correspondence.
[0087] Understandably, the baseline distance is calculated by substituting the sample distance between the pre-trained model parameters, the baseline distance coefficient, and the number of pre-trained model parameters into the first correspondence mentioned above.
[0088] Step S2013: Determine the close parameters corresponding to the pre-trained model parameters based on the sample distance and the baseline distance. The sample distance between the pre-trained model parameters and the corresponding close parameters is less than the baseline distance.
[0089] It should be noted that close parameters refer to parameters that are close to the pre-trained model parameters. In this embodiment, the sample distance between the pre-trained model parameters and the corresponding close parameters needs to be less than the baseline distance, so that the close parameters corresponding to each pre-trained model parameter can be determined.
[0090] Step S2014: Obtain the second correspondence between the number of tight parameters, the number of pre-trained model parameters, and the sample tightness;
[0091] It should be noted that the number of tight parameters refers to the number of tight parameters in each pre-trained model. The second correspondence between the number of tight parameters, the number of pre-trained model parameters, and sample tightness, i.e., the formula for calculating sample tightness, is as follows:
[0092]
[0093] In the formula, TI represents the sample density. This represents the number of closely spaced parameters, where n represents the number of parameters in the pre-trained model.
[0094] Step S2015: Based on the number of closely related parameters of the pre-trained model parameters, the number of pre-trained model parameters, and the second correspondence, the sample density of the pre-trained model parameters is obtained.
[0095] It is understandable that by substituting the number of closely related parameters of the pre-trained model parameters and the number of pre-trained model parameters into the second correspondence mentioned above, the sample density of each pre-trained model parameter can be calculated.
[0096] Step S202: Based on the baseline distance and sample density of the pre-trained model parameters, the pre-trained model parameters are divided into corresponding sample clusters;
[0097] In one possible implementation, step S202 may include steps S2021 to S2026:
[0098] Step S2021: Based on the sample density of the pre-trained model parameters, sort the pre-trained model parameters in descending order to obtain the parameter sequence, and add the first pre-trained model parameter in the parameter sequence as the initial cluster center point to the center point set;
[0099] It should be noted that the sequence of pre-trained model parameters, arranged from highest to lowest sample density, is the parameter sequence. The initial cluster centroids refer to the initially determined cluster centroids, which will require further adjustment later. The first pre-trained model parameter in the parameter sequence is the first signal in the sequence, representing the pre-trained model parameter with the highest sample density; this parameter is the first selected initial cluster centroid by default. The centroid set is the set of initial cluster centroids, into which all initial cluster centroids are added.
[0100] Understandably, based on the sample density of each pre-trained model parameter, all vibration signals are arranged from largest to smallest to form a parameter sequence P = {P1, P2, ..., P...}. n}, and construct a set of center points C, with P1 automatically selected as the first initial cluster center point and added to the set of center points C.
[0101] Step S2022: Determine the center point screening threshold based on the cluster center difference parameter and the baseline distance;
[0102] It should be noted that the center point screening threshold refers to the distance threshold set when selecting the initial cluster center point, while the cluster center difference parameter refers to the relevant parameter that needs to be considered when calculating the center point screening threshold. Its initial value is 1, and it can be flexibly adjusted according to actual needs. The center point screening threshold is obtained by multiplying the cluster center difference parameter by the baseline distance.
[0103] Step S2023: Obtain the center distance between the pre-trained model parameters and the initial cluster center points in the center point set; add the pre-trained model parameters whose center distance is greater than the center point screening threshold as the initial cluster center points to the center point set.
[0104] It should be noted that the center distance refers to the distance between the pre-trained model parameters and all selected initial cluster center points.
[0105] It is understandable that if the center distances between the parameters of a pre-trained model and the initial cluster centers in the set of center points all satisfy... in, This represents the pre-trained model parameters of client i. With the initial cluster center point c j The center distance between them, μ represents the cluster center difference parameter, L B This represents the baseline distance. At this point, the pre-trained model parameter can be used as the initial cluster center point. In other words, if the distance between the pre-trained model parameter and each initial cluster center point in the center point set is greater than the center point selection threshold, the pre-trained model parameter is considered as the next initial cluster center point and added to the center point set.
[0106] Step S2024: When the number of initial cluster center points in the center point set reaches a preset number, based on the center distance between the pre-trained model parameters and the initial cluster center points, determine the cluster center points corresponding to the pre-trained model parameters, divide the pre-trained model parameters into sample clusters of the corresponding cluster center points, and determine the distance between the pre-trained model parameters and the cluster center points in the sample clusters.
[0107] It should be noted that the preset number is the number of initial cluster centers to be selected, which is the number of sample clusters to be divided. This preset number can be set according to the actual situation, and there is no specific restriction on it. If the number of initial cluster centers in the center point set reaches the preset number, it means that all initial cluster centers have been selected. If the number of initial cluster centers in the center point set does not reach the preset number, step S2023 needs to be executed again to select initial cluster centers.
[0108] Understandably, each pre-trained model parameter is usually assigned to the cluster containing the nearest cluster center. In other words, the cluster center corresponding to the pre-trained model parameter is the initial cluster center closest to that pre-trained model parameter. At this time, the pre-trained model parameter needs to be assigned to the sample cluster containing that cluster center, and the distance between each pre-trained model parameter in the sample cluster and the cluster center is calculated.
[0109] Step S2025: Based on the distance between the pre-trained model parameters in the sample cluster and the cluster center point, determine the mean model parameters of the sample cluster, and use the mean model parameters of the sample cluster as the new cluster center point.
[0110] It should be noted that the mean model parameter refers to the mean of all pre-trained model parameters in the sample cluster. After dividing all the pre-trained model parameters, the mean model parameter of the sample cluster is used as the new cluster centroid.
[0111] Step S2026: When the clustering conditions are met, determine the sample clusters corresponding to the pre-trained model parameters based on the distance between the pre-trained model parameters and the cluster center points, and execute the step of determining the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters based on the sample clusters of the pre-trained model parameters; when the clustering conditions are not met, return to the step of determining the distance between the pre-trained model parameters and the cluster center points in the sample clusters.
[0112] It is understood that the clustering conditions, i.e. the conditions for completing clustering, can be set according to actual needs, and this implementation does not impose specific restrictions on them. If the clustering conditions are met, the cluster center points obtained at this time are the final determined cluster center points. According to the cluster center points, all pre-trained model parameters are re-divided, and each pre-trained model parameter is assigned to the sample cluster where the nearest cluster center point is located. Step S203 is then executed. If the clustering conditions are not met, the cluster center points obtained at this time need to be further adjusted, the distance between the pre-trained model parameters and the cluster center points is recalculated, and the cluster center points are re-determined.
[0113] Step S203: Based on the sample clusters of the pre-trained model parameters, determine the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters.
[0114] Understandably, the central server records the category of the engine misfire fault identification model corresponding to each pre-trained model parameter according to the sample cluster to which each pre-trained model parameter belongs, and then sends it to the corresponding client.
[0115] In one feasible implementation, engine misfire fault identification models of the same category can be aggregated into a central model.
[0116] In this embodiment, participating clients are categorized to help them discover similarities and correlations between data, promote knowledge sharing and cooperation among participants, and improve the recognition accuracy of the final engine misfire fault identification model.
[0117] Step S30: Based on the local high-quality vibration signal, continue to train the engine misfire fault identification model, obtain the training model parameters of the engine misfire fault identification model uploaded by the client, and determine the client's aggregation influence factor based on the training model parameters and aggregation category.
[0118] It should be noted that the client uses local high-quality vibration signals to train the engine misfire fault identification model. After the initial training is completed, each client uploads its current model parameters to the central server. The trained model parameters are the model parameters of the engine misfire fault identification model trained on the client. The aggregation impact factor can characterize the difference between the model parameters of each client and the model parameters aggregated by the central server. It is equivalent to evaluating the degree of benefit / advantage of each client to the global model, so as to better utilize the clients that are beneficial to the global model update, which is conducive to subsequent parameter aggregation.
[0119] In one feasible implementation, the step of determining the client's aggregation influence factor based on the training model parameters and aggregation category may include steps S301 to S303:
[0120] Step S301: Based on the aggregation category of the engine misfire fault identification model in the client, determine the cluster center point of the corresponding sample cluster in the client, and determine the aggregation distance between the training model parameters of the client and the cluster center point.
[0121] It should be noted that, based on the aggregation category of the engine misfire fault identification model in the client, the cluster center point of the corresponding sample cluster in the client is found. The aggregation distance between the client's training model parameters and the cluster center point refers to the distance between the client's training model parameters and the cluster center point of the corresponding sample cluster in the client. The calculation relationship is as follows:
[0122]
[0123] In the formula, W represents the training model parameters of client i. c This indicates the cluster center point of the sample cluster corresponding to client i. Indicates the aggregation distance.
[0124] Step S302: Obtain the third correspondence between aggregation distance and aggregation influence factor;
[0125] It should be noted that the third correspondence between aggregation distance and aggregation impact factor, i.e., the calculation formula for aggregation impact factor, is as follows:
[0126]
[0127] In the formula, σ i This represents the aggregate influence factor of client i. Indicates the aggregation distance.
[0128] Step S303: Determine the aggregation influence factor of the client based on the aggregation distance corresponding to the client and the third correspondence.
[0129] Understandably, by substituting the aggregation distances corresponding to each client into the third correspondence mentioned above, the aggregation impact factor for each client can be calculated.
[0130] Step S40: Based on the client's aggregation influence factor, aggregate the training model parameters of clients in the same aggregation category to obtain the aggregation model parameters corresponding to each aggregation category;
[0131] In one feasible implementation, step S40 may include steps S401 to S402:
[0132] Step S401: Obtain the fourth correspondence between the aggregated influence factor, training model parameters, number of signals from the client, total number of signals, and aggregated model parameters;
[0133] It should be noted that the aggregated model parameters refer to the model parameters obtained by aggregating the training model parameters of each client within the same aggregation category; that is, the global model parameters. The number of signals from each client refers to the number of high-quality local vibration signals from each client, and the total number of signals refers to the total number of high-quality local vibration signals from all clients participating in the aggregation. The fourth correspondence between the aggregation influence factor, training model parameters, number of signals from each client, total number of signals, and aggregated model parameters refers to the calculation formula for the aggregated model parameters, as shown below:
[0134]
[0135] Right now:
[0136]
[0137] In the formula, Represents the parameters of the aggregation model. X represents the training model parameters of client i. i σ represents the number of signals from client i, X represents the total number of signals in this round, and σ represents the total number of signals from client i. i This represents the aggregate impact factor of client i in this round. This represents the aggregation distance of client i in this round.
[0138] Step S402: Determine the aggregation model parameters corresponding to each aggregation category based on the client's aggregation influence factor, the training model parameters of the engine misfire fault identification model, the number of signals from the client, the total number of signals, and the fourth correspondence.
[0139] Understandably, by substituting the client's aggregation influence factor, the training model parameters of the engine misfire fault identification model, the number of client signals, and the total number of signals into the second correspondence mentioned above, the training model parameters of clients in the same aggregation category are aggregated, and the aggregated model parameters corresponding to each aggregation category are calculated.
[0140] It should be understood that the aggregation impact factor for each round is dynamically calculated based on the training model parameters of each client, and the aggregation impact factor is used to calculate the aggregated model parameters, thereby realizing the autonomous dynamic aggregation of the training model parameters.
[0141] Step S50: The aggregated model parameters are sent to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model. Based on the target engine misfire fault identification model, misfire fault monitoring of the engine is performed.
[0142] Understandably, after obtaining the aggregated model parameters, the central server distributes the aggregated model parameters to each client. Each client updates its local engine misfire fault identification model and then continues to perform the next round of model updates until the target engine misfire fault identification model is obtained. The target engine misfire fault identification model is the final engine misfire fault identification model.
[0143] In one feasible implementation, the step of sending aggregation model parameters to the client based on the aggregation category corresponding to the client, so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model, may include: sending aggregation model parameters to the client so that the client can update the engine misfire fault identification model and determining whether the iteration termination condition is met; if the iteration termination condition is met, using the updated engine misfire fault identification model as the target engine misfire fault identification model, and performing the step of monitoring engine misfire fault based on the target engine misfire fault identification model; if the iteration termination condition is not met, retraining the engine misfire fault identification model based on local high-quality vibration signals, updating the training model parameters, and returning to the step of obtaining the training model parameters of the engine misfire fault identification model uploaded by the client, and determining the client's aggregation influence factor based on the training model parameters and the aggregation category.
[0144] It should be noted that the iteration termination condition refers to the termination condition of model training, which can be considered as the termination condition of federated model training, such as: the number of iterations reaching a set threshold. This embodiment does not make specific limitations on this.
[0145] Understandably, the corresponding aggregation model parameters are sent to each client according to the aggregation category of the client. After the client updates the engine misfire fault identification model using the aggregation model parameters, if the iteration termination condition is met, it means that the model iteration update is complete, and the local model learning on the client ends. At this time, the final engine misfire fault identification model can be obtained, and the client can use the final engine misfire fault identification model to identify the current misfire fault mode of the engine. If the iteration termination condition is not met, it means that the model iteration update is not complete, and the model continues to be iteratively trained, the training model parameters uploaded by the client are updated, and the process returns to step S30 to recalculate the aggregation influence factor, thereby redetermining the aggregation model parameters and updating the engine misfire fault identification model again.
[0146] It should be understood that different misfire failure modes correspond to different optimization strategies. Optimization strategies are suggestions / measures for optimizing misfire failures, and can be used to optimize the engine.
[0147] It should be understood that by using the engine misfire fault identification model to determine the engine misfire fault mode in real time, locate the misfired cylinder, and adopt corresponding optimization suggestions / measures to optimize the engine, the vehicle's performance can be optimized, thus ensuring the vehicle's performance and safety.
[0148] This embodiment provides a method for engine misfire fault identification under data security collaboration. Initial model parameters are distributed to multiple clients, enabling each client to initialize an engine misfire fault identification model based on these parameters. The model is then pre-trained using local high-quality vibration signals. Pre-trained model parameters from client-uploaded models are acquired, and the models are categorized to determine their aggregation categories. The model is further trained using local high-quality vibration signals, and training model parameters from client-uploaded models are acquired. Aggregation influence factors for each client are determined based on these training model parameters and the aggregation categories. The training model parameters from clients within the same aggregation category are aggregated using these aggregation influence factors to obtain aggregated model parameters for each category. These aggregated model parameters are then distributed to the corresponding clients, allowing them to iteratively update their engine misfire fault identification models to obtain a target model. Finally, engine misfire fault monitoring is performed based on this target model. Each client utilizes its local high-quality data to build a local model. After aggregating the model parameters from each client, the resulting aggregated model parameters are distributed to each client for iterative model updates to obtain the final engine misfire fault identification model. During this process, by sharing model parameters from different clients—that is, sharing high-quality data while ensuring data privacy and security—data quality is enhanced, thereby improving the identification accuracy of the engine misfire fault identification model. Furthermore, the aggregation strategy can be changed in real time according to actual conditions, effectively filtering and utilizing clients beneficial to global model updates, further improving model performance and identification accuracy. In addition, classifying all participating clients helps participants discover similarities and correlations between data, promoting knowledge sharing and cooperation among participants, further improving the model's identification accuracy.
[0149] For example, to help understand the implementation process of the engine misfire fault identification method under data security integration obtained in this embodiment combined with the above embodiment one, please refer to... Figure 2 , Figure 2 A simplified architectural diagram of an engine misfire fault identification method under data security collaboration is provided, specifically:
[0150] The overall architecture includes at least n clients and a central server trusted by all clients. Clients upload the parameters of their pre-trained local models. The central server classifies all clients' local models based on these parameters, constructs a central model with multiple categories, and distributes the category information. Clients train their local models and upload their parameters. The central server calculates aggregation metrics (e.g., aggregation impact factors) based on the parameters of each client's local model, aggregates the parameters of each client's model based on these metrics, and distributes the aggregated parameters to each client. Each client iteratively updates its local model based on the aggregated parameters to obtain the final engine misfire fault identification model, used for analyzing and identifying engine misfire faults. A pre-clustering module can be independently set up to classify the clients' local models and construct a central model with multiple categories. Alternatively, a client evaluation strategy calculation unit can be independently set up to calculate the parameters for each round of aggregation and store the parameters for the current round of aggregation.
[0151] Please refer to Figure 3 , Figure 3 A simplified flowchart of an engine misfire fault identification method under data security collaboration is provided, specifically:
[0152] Based on training requirements, the central server distributes the initial convolutional neural network model structure to n clients. Each client initializes its local engine misfire fault recognition model. The clients use local high-quality vibration signals and corresponding misfire fault mode labels (such as single-cylinder single misfire, single-cylinder continuous misfire, multi-cylinder single misfire, and multi-cylinder continuous misfire) as training data to pre-train their local engine misfire fault recognition models. Each client uploads the pre-trained model parameters to the central server. The central server classifies all client models, constructs multiple categories of central models (such as three-cylinder, four-cylinder, six-cylinder, etc.), and distributes the category information to each client. Each client then begins utilizing its local high-quality vibration signals... The engine misfire identification model is trained using vibration signals and corresponding misfire fault mode labels. After the initial training of the local model, each client uploads the parameters of its local model to the central server. The central server calculates the aggregation influence factor of each client in this overall iteration based on the parameters of the local model uploaded by each client. The central server aggregates the parameters of the local model of each client according to the autonomous dynamic aggregation strategy based on the aggregation influence factor, and distributes the aggregated parameters to each client. Each client updates its local model and starts a new round of local model training until the training termination condition is met, ending the local federated model learning and obtaining the final engine misfire fault identification model.
[0153] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the engine misfire fault identification method under the data security joint of this application. Any simple modifications based on this technical concept are within the protection scope of this application.
[0154] This application also provides an engine misfire fault identification device under data security integration. Please refer to... Figure 4 The data security-enabled engine misfire fault identification device includes:
[0155] The model training module 10 is used to distribute the initial model parameters to multiple clients so that the clients can initialize the engine misfire fault identification model according to the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals.
[0156] The model pre-clustering module 20 is used to obtain the pre-trained model parameters of the engine misfire fault identification model uploaded by the client, classify the engine misfire fault identification model based on the pre-trained model parameters, and determine the aggregation category of the engine misfire fault identification model in the client.
[0157] The aggregation factor calculation module 30 is used to continue training the engine misfire fault identification model based on local high-quality vibration signals, obtain the training model parameters of the engine misfire fault identification model uploaded by the client, and determine the aggregation influence factor of the client based on the training model parameters and aggregation category.
[0158] The parameter aggregation module 40 is used to aggregate the training model parameters of clients in the same aggregation category based on the client's aggregation influence factor, so as to obtain the aggregated model parameters corresponding to each aggregation category.
[0159] The misfire fault identification module 50 is used to send the aggregated model parameters to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model. Based on the target engine misfire fault identification model, the engine is monitored for misfire faults.
[0160] In one embodiment, the model pre-clustering module 20 is further configured to determine the baseline distance and sample compactness of the pre-trained model parameters based on the sample distance between the pre-trained model parameters;
[0161] Based on the baseline distance and sample density of the pre-trained model parameters, the pre-trained model parameters are divided into corresponding sample clusters;
[0162] Based on the sample clusters of pre-trained model parameters, the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters is determined.
[0163] In one embodiment, the model pre-clustering module 20 is further configured to obtain a first correspondence between the sample distance, the baseline distance coefficient, the number of pre-trained model parameters, and the baseline distance;
[0164] The baseline distance is determined based on the sample distance between pre-trained model parameters, the baseline distance coefficient, the number of pre-trained model parameters, and the first correspondence.
[0165] Based on the sample distance and the baseline distance, determine the close parameters corresponding to the pre-trained model parameters. The sample distance between the pre-trained model parameters and the corresponding close parameters is less than the baseline distance.
[0166] To obtain a second correspondence between the number of tight parameters, the number of pre-trained model parameters, and sample tightness;
[0167] The sample density of the pre-trained model parameters is obtained based on the number of closely corresponding parameters of the pre-trained model parameters, the number of pre-trained model parameters, and the second correspondence.
[0168] In one embodiment, the model pre-clustering module 20 is further configured to sort the pre-trained model parameters in descending order according to the sample density of the pre-trained model parameters to obtain a parameter sequence, and add the first pre-trained model parameter in the parameter sequence as the initial cluster center point to the center point set;
[0169] The center point screening threshold is determined based on the cluster center difference parameter and the baseline distance;
[0170] Obtain the center distance between the pre-trained model parameters and the initial cluster center points in the center point set. Select the pre-trained model parameters whose center distance is greater than the center point screening threshold as the initial cluster center points and add them to the center point set.
[0171] When the number of initial cluster center points in the center point set reaches a preset number, the cluster center point corresponding to the pre-trained model parameters is determined based on the center distance between the pre-trained model parameters and the initial cluster center points. The pre-trained model parameters are then assigned to the sample clusters of the corresponding cluster center points, and the distance between the pre-trained model parameters and the cluster center points in the sample clusters is determined.
[0172] Based on the distance between the pre-trained model parameters in the sample cluster and the cluster center point, the mean model parameters of the sample cluster are determined, and the mean model parameters of the sample cluster are used as the new cluster center point.
[0173] When the clustering conditions are met, the sample clusters corresponding to the pre-trained model parameters are determined based on the distance between the pre-trained model parameters and the cluster center points. Then, the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters is determined based on the sample clusters of the pre-trained model parameters.
[0174] If the clustering conditions are not met, return to the step of determining the distance between the pre-trained model parameters and the cluster centroid in the sample cluster.
[0175] In one embodiment, the aggregation factor calculation module 30 is further configured to determine the cluster center point of the corresponding sample cluster of the client based on the aggregation category of the engine misfire fault identification model in the client, and to determine the aggregation distance between the training model parameters of the client and the cluster center point.
[0176] To obtain the third correspondence between aggregation distance and aggregation influence factor;
[0177] The aggregation influence factor of the client is determined based on the aggregation distance corresponding to the client and the third correspondence.
[0178] In one embodiment, the parameter aggregation module 40 is further configured to obtain a fourth correspondence between the aggregation influence factor, training model parameters, number of signals from the client, total number of signals, and aggregation model parameters;
[0179] Based on the client's aggregation influence factor, the training model parameters of the engine misfire fault identification model, the number of signals from the client, the total number of signals, and the fourth correspondence, the aggregation model parameters corresponding to each aggregation category are determined.
[0180] In one embodiment, the misfire fault identification module 50 is further configured to send the aggregate model parameters to the client based on the aggregate category corresponding to the client, so that the client updates the engine misfire fault identification model and determines whether the iteration termination condition is met.
[0181] When the iteration termination condition is met, the updated engine misfire fault identification model is used as the target engine misfire fault identification model, and the steps of monitoring engine misfire fault based on the target engine misfire fault identification model are executed.
[0182] If the iteration termination condition is not met, the engine misfire fault identification model is retrained based on the local high-quality vibration signal, and the training model parameters are updated. Then, the process returns to retrieve the training model parameters of the engine misfire fault identification model uploaded by the client, and determines the client's aggregation influence factor based on the training model parameters and aggregation category.
[0183] In one embodiment, the client's local high-quality vibration signal is provided with a corresponding training label. The training label is set according to the misfire fault mode of the local high-quality vibration signal. The misfire fault mode includes at least single-cylinder single misfire, single-cylinder continuous misfire, multi-cylinder single misfire, and multi-cylinder continuous misfire.
[0184] In one embodiment, different misfire failure modes correspond to different optimization strategies, which are used to optimize the engine.
[0185] The engine misfire fault identification device under data security integration provided in this application, employing the engine misfire fault identification method under data security integration described in the above embodiments, can solve the technical problem that traditional methods are unable to accurately determine the engine misfire fault mode. Compared with the prior art, the beneficial effects of the engine misfire fault identification device under data security integration provided in this application are the same as those of the engine misfire fault identification method under data security integration provided in the above embodiments, and other technical features in the engine misfire fault identification device under data security integration are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.
[0186] This application provides an engine misfire fault identification device under data security integration. The engine misfire fault identification device under data security integration includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the engine misfire fault identification method under data security integration described in Embodiment 1 above.
[0187] The following is for reference. Figure 5 This document illustrates a structural schematic diagram of an engine misfire fault identification device suitable for implementing the data security-enabled embodiments of this application. The engine misfire fault identification device under data security-enabled implementation in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 5 The engine misfire fault identification device shown under the data security alliance is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0188] like Figure 5As shown, the engine misfire fault detection device under data security integration may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the engine misfire fault detection device under data security integration. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the engine misfire fault detection device under data security integration to exchange data with other devices wirelessly or via wired communication. Although the figure shows an engine misfire fault detection device under data security integration with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.
[0189] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application 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 a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.
[0190] The engine misfire fault identification device under data security integration provided in this application, employing the engine misfire fault identification method under data security integration described in the above embodiments, can solve the technical problem that traditional methods are unable to accurately determine the engine misfire fault mode. Compared with the prior art, the beneficial effects of the engine misfire fault identification device under data security integration provided in this application are the same as those of the engine misfire fault identification method under data security integration provided in the above embodiments, and other technical features in this engine misfire fault identification device under data security integration are the same as those disclosed in the previous embodiment method, and will not be repeated here.
[0191] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.
[0192] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0193] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the engine misfire fault identification method under data security coordination in the above embodiments.
[0194] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.
[0195] The aforementioned computer-readable storage medium may be included in the engine misfire fault identification device under data security integration; or it may exist independently and not be assembled into the engine misfire fault identification device under data security integration.
[0196] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the engine misfire fault identification device under data security collaboration, the engine misfire fault identification device under data security collaboration: distributes initial model parameters to multiple clients, enabling the clients to initialize engine misfire fault identification models based on the initial model parameters and pre-train engine misfire fault identification models based on local high-quality vibration signals; acquires the pre-trained model parameters of the engine misfire fault identification models uploaded by the clients; classifies the engine misfire fault identification models based on the pre-trained model parameters; and determines the clustering of engine misfire fault identification models in the clients. The system is categorized and, based on high-quality local vibration signals, continues to train the engine misfire fault identification model. It acquires the training model parameters uploaded by the client and determines the client's aggregation influence factor based on these parameters and the aggregation category. Based on the client's aggregation influence factor, it aggregates the training model parameters of clients within the same aggregation category to obtain the aggregated model parameters corresponding to each category. The aggregated model parameters are then distributed to the corresponding clients to enable iterative updates of the engine misfire fault identification model, resulting in the target engine misfire fault identification model. Finally, based on the target engine misfire fault identification model, the system monitors for engine misfire faults.
[0197] Computer program code for performing the operations of this application can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as the "C" language or similar 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 a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0198] 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 this application. 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 the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can 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.
[0199] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.
[0200] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the engine misfire fault identification method under the aforementioned data security integration, thereby solving the technical problem that traditional methods are unable to accurately determine the engine misfire fault mode. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the engine misfire fault identification method under the aforementioned data security integration, and will not be repeated here.
[0201] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the engine misfire fault identification method under the aforementioned data security integration.
[0202] The computer program product provided in this application can solve the technical problem that traditional methods are unable to accurately determine the misfire failure mode of an engine. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the engine misfire failure identification method under data security integration provided in the above embodiments, and will not be repeated here.
[0203] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.
Claims
1. A method for identifying engine misfire faults under data security collaboration, characterized in that, The method includes: The initial model parameters are distributed to multiple clients so that the clients initialize the engine misfire fault identification model according to the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals. Obtain the pre-trained model parameters of the engine misfire fault identification model uploaded by the client, classify the engine misfire fault identification model based on the pre-trained model parameters, and determine the aggregate category of the engine misfire fault identification model in the client. Based on local high-quality vibration signals, the engine misfire fault identification model is further trained, the training model parameters of the engine misfire fault identification model uploaded by the client are obtained, and the aggregation influence factor of the client is determined based on the training model parameters and the aggregation category. Based on the aggregation influence factor of the client, the training model parameters of clients in the same aggregation category are aggregated to obtain the aggregation model parameters corresponding to each aggregation category. The aggregated model parameters are sent to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model. Based on the target engine misfire fault identification model, misfire fault monitoring of the engine is performed. The step of classifying the engine misfire fault identification model based on the pre-trained model parameters and determining the aggregation category of the engine misfire fault identification model includes: obtaining a first correspondence between sample distance, baseline distance coefficient, the number of pre-trained model parameters, and the baseline distance; determining the baseline distance based on the sample distance between the pre-trained model parameters, the baseline distance coefficient, the number of pre-trained model parameters, and the first correspondence; determining the compact parameters corresponding to the pre-trained model parameters based on the sample distance and the baseline distance, wherein the sample distance between the pre-trained model parameters and the corresponding compact parameters is less than the baseline distance; obtaining a second correspondence between the number of compact parameters, the number of pre-trained model parameters, and the sample compactness; obtaining the sample compactness of the pre-trained model parameters based on the number of compact parameters corresponding to the pre-trained model parameters, the number of pre-trained model parameters, and the second correspondence; dividing the pre-trained model parameters into corresponding sample clusters based on the baseline distance and sample compactness; and determining the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters based on the sample clusters of the pre-trained model parameters. The first correspondence is The second correspondence is , Indicates the reference distance. Indicates the reference distance coefficient. For pre-trained model parameters and The sample distance between them This indicates the number of parameters in the pre-trained model. Indicates sample density. Indicates the number of tight parameters.
2. The method as described in claim 1, characterized in that, The step of dividing the pre-trained model parameters into corresponding sample clusters based on the baseline distance and sample density of the pre-trained model parameters includes: Based on the sample density of the pre-trained model parameters, the pre-trained model parameters are sorted in descending order to obtain a parameter sequence. The first pre-trained model parameter in the parameter sequence is added to the center point set as the initial cluster center point. The center point screening threshold is determined based on the cluster center difference parameter and the benchmark distance; Obtain the center distance between the pre-trained model parameters and the initial cluster center points in the center point set; add the pre-trained model parameters whose center distance is greater than the center point screening threshold as initial cluster center points to the center point set. When the number of initial cluster centers in the set of centers reaches a preset number, based on the center distance between the pre-trained model parameters and the initial cluster centers, the cluster center corresponding to the pre-trained model parameters is determined, and the pre-trained model parameters are assigned to the sample clusters of the corresponding cluster centers, and the distance between the pre-trained model parameters and the cluster centers in the sample clusters is determined. Based on the distance between the pre-trained model parameters in the sample cluster and the cluster center point, the mean model parameters of the sample cluster are determined, and the mean model parameters of the sample cluster are used as the new cluster center point. When the clustering conditions are met, the sample clusters corresponding to the pre-trained model parameters are determined based on the distance between the pre-trained model parameters and the cluster center point. Then, the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters is determined based on the sample clusters of the pre-trained model parameters. If the clustering conditions are not met, return to the step of determining the distance between the pre-trained model parameters and the cluster centroid in the sample cluster.
3. The method as described in claim 1, characterized in that, The step of determining the aggregation influence factor of the client based on the training model parameters and the aggregation category includes: Based on the aggregation category of the engine misfire fault identification model in the client, the cluster center point of the corresponding sample cluster of the client is determined, and the aggregation distance between the training model parameters of the client and the cluster center point is determined. To obtain the third correspondence between aggregation distance and aggregation influence factor; The aggregation influence factor of the client is determined based on the aggregation distance corresponding to the client and the third correspondence.
4. The method as described in claim 1, characterized in that, The step of aggregating the training model parameters of clients in the same aggregation category based on the aggregation influence factor of the client to obtain the aggregation model parameters corresponding to each aggregation category includes: Obtain the fourth correspondence between the aggregated impact factor, training model parameters, number of signals from clients, total number of signals, and aggregated model parameters; Based on the client's aggregation influence factor, the training model parameters of the engine misfire fault identification model, the number of signals from the client, the total number of signals, and the fourth correspondence, the aggregation model parameters corresponding to each aggregation category are determined.
5. The method as described in claim 1, characterized in that, The step of sending the aggregated model parameters to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model includes: Based on the aggregation category corresponding to the client, the aggregation model parameters are sent to the client so that the client can update the engine misfire fault identification model and determine whether the iteration termination condition is met. When the iteration termination condition is met, the updated engine misfire fault identification model is used as the target engine misfire fault identification model, and the step of monitoring engine misfire fault based on the target engine misfire fault identification model is executed. If the iteration termination condition is not met, the engine misfire fault identification model is retrained based on the local high-quality vibration signal, and the training model parameters are updated. Then, the process returns to the step of obtaining the training model parameters of the engine misfire fault identification model uploaded by the client, and determining the aggregation influence factor of the client based on the training model parameters and the aggregation category.
6. The method according to any one of claims 1 to 5, characterized in that, The client's local high-quality vibration signal is equipped with a corresponding training label. The training label is set according to the misfire failure mode of the local high-quality vibration signal. The misfire failure mode includes at least single-cylinder single misfire, single-cylinder continuous misfire, multi-cylinder single misfire, and multi-cylinder continuous misfire. Different misfire failure modes correspond to different optimization strategies. The optimization strategies are used to optimize the engine.
7. A data security-enabled engine misfire fault identification device, characterized in that, The engine misfire fault identification device under the data security joint system includes: The model training module is used to send initial model parameters to multiple clients so that the clients initialize the engine misfire fault identification model according to the initial model parameters and pre-train the engine misfire fault identification model based on local high-quality vibration signals. The model pre-clustering module is used to obtain the pre-trained model parameters of the engine misfire fault identification model uploaded by the client, classify the engine misfire fault identification model based on the pre-trained model parameters, and determine the aggregation category of the engine misfire fault identification model in the client. The aggregation factor calculation module is used to continue training the engine misfire fault identification model based on the local high-quality vibration signal, obtain the training model parameters of the engine misfire fault identification model uploaded by the client, and determine the aggregation influence factor of the client based on the training model parameters and the aggregation category. The parameter aggregation module is used to aggregate the training model parameters of clients in the same aggregation category based on the aggregation influence factor of the client, so as to obtain the aggregation model parameters corresponding to each aggregation category. The misfire fault identification module is used to send the aggregated model parameters to the corresponding client so that the client can iteratively update the engine misfire fault identification model to obtain the target engine misfire fault identification model, and monitor the engine for misfire fault based on the target engine misfire fault identification model. The model pre-clustering module is also used to obtain the first correspondence between the sample distance, the baseline distance coefficient, the number of pre-trained model parameters and the baseline distance; The baseline distance is determined based on the sample distance between the pre-trained model parameters, the baseline distance coefficient, the number of pre-trained model parameters, and the first correspondence. Based on the sample distance and the baseline distance, the compact parameters corresponding to the pre-trained model parameters are determined, wherein the sample distance between the pre-trained model parameters and the corresponding compact parameters is less than the baseline distance; To obtain a second correspondence between the number of tight parameters, the number of pre-trained model parameters, and sample tightness; The sample density of the pre-trained model parameters is obtained based on the number of closely related parameters to the pre-trained model parameters, the number of pre-trained model parameters, and the second correspondence. Based on the baseline distance and sample density of the pre-trained model parameters, the pre-trained model parameters are divided into corresponding sample clusters; Based on the sample clusters of the pre-trained model parameters, determine the aggregation category of the engine misfire fault identification model corresponding to the pre-trained model parameters; The first correspondence is The second correspondence is , Indicates the reference distance. Indicates the reference distance coefficient. For pre-trained model parameters and The sample distance between them This indicates the number of parameters in the pre-trained model. Indicates sample density. Indicates the number of tight parameters.
8. An engine misfire fault identification device under data security coordination, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the engine misfire fault identification method under data security as described in any one of claims 1 to 6.
Citation Information
Patent Citations
Method for distributing participant weights in federated learning based on clustering algorithm
CN114758784A
