Method and system for engine test platform fault early warning based on time sequence mining

Abstract

The application discloses a timing mining-based engine test platform fault early warning method and system, and relates to the field of engine testing, wherein the method comprises the following steps: obtaining test control instructions, engine body and test platform monitoring data of the engine during the test process; performing response deviation detection on the monitoring data according to the test control instructions, and establishing a candidate abnormal variable space; constructing a cross-domain variable propagation topology between the engine body and the test platform, performing abnormal occurrence timing tracking on the candidate abnormal variable space, and generating a cross-domain abnormal propagation chain; calculating the fault propagation matching degree of the engine body and the test platform; performing safety verification disturbance testing according to the matching degree, obtaining a test bilateral response sequence; locating the fault source, and generating a fault early warning instruction according to the fault source location result. The application solves the technical problem that the existing engine test fault early warning has insufficient early warning directivity due to unknown monitoring signal source attribution.

Description

Technical Field

[0001] This application relates to the field of engine testing, and in particular to a fault early warning method and system for engine testing platforms based on time-series mining. Background Technology

[0002] Fault warning during engine testing is crucial for ensuring test safety and the validity of results. Current mainstream technologies primarily rely on data acquisition systems and monitoring software platforms. By setting fixed thresholds for various monitoring parameters or using statistical process control methods, they assess real-time test data for exceeding limits and monitoring trends, triggering alarms when parameters exceed preset thresholds. However, due to the strong coupling between the engine and the test platform in terms of physical structure and signal response, their monitoring characteristics under abnormal conditions are often highly similar. Existing methods struggle to effectively distinguish the actual source of the anomaly from the coupled monitoring signals, resulting in a high false alarm rate and ambiguous localization in the warning results.

[0003] At present, the engine test fault warning technology suffers from a technical problem: the source of the monitoring signal is unclear, resulting in insufficient warning directionality. Summary of the Invention

[0004] This application provides a fault early warning method and system for engine test platforms based on time-series mining. By acquiring three types of raw data during the testing process, it performs response deviation detection on two types of monitoring data to screen for abnormal variables, constructs a cross-domain topology and performs time-series tracking of abnormal variables to generate propagation chains, calculates the propagation matching degree under two fault assumptions, performs safety disturbance testing and collects bilateral responses when the matching degree is close, and finally locates the fault source and outputs early warning commands based on the response data. These technical means solve the technical problem of insufficient early warning directionality caused by unclear monitoring signal source attribution in existing engine test fault early warning systems. It achieves the technical effect of accurately locating the fault source and improving the directionality and effectiveness of early warning commands under the condition of mutual coupling and interference between engine body monitoring signals and test platform monitoring signals.

[0005] This application provides a fault early warning method for an engine test platform based on time-series mining, comprising: acquiring test control commands, engine body monitoring data, and test platform monitoring data during engine testing; performing response deviation detection on the engine body monitoring data and the test platform monitoring data according to the test control commands, and establishing a candidate abnormal variable space; constructing a cross-domain variable propagation topology between the engine body and the test platform, and performing anomaly occurrence time-series tracking on the candidate abnormal variable space according to the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain; calculating the engine body fault propagation matching degree and the test platform fault propagation matching degree according to the cross-domain anomaly propagation chain; performing a safety verification disturbance test according to the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a bilateral test response sequence; locating the fault source according to the bilateral test response sequence, obtaining the fault source location result, and generating a fault early warning command according to the fault source location result.

[0006] In a possible implementation, response deviation detection is performed on the engine body monitoring data and the test platform monitoring data according to the test control instructions, respectively, to establish a candidate anomaly variable space, and the following processing is performed: bilateral response baseline mining is performed according to the test control instructions to establish an engine response baseline feature matrix and a platform response baseline feature matrix; response deviation detection is performed on the engine body monitoring data according to the engine response baseline feature matrix to obtain an engine response deviation sequence; response deviation detection is performed on the test platform monitoring data according to the platform response baseline feature matrix to obtain a platform response deviation sequence; the engine response deviation sequence and the platform response deviation sequence are filtered according to anomaly screening conditions to generate the candidate anomaly variable space.

[0007] In a possible implementation, bilateral response baseline mining is performed according to the test control command to establish an engine response baseline feature matrix and a platform response baseline feature matrix. The following processes are then performed: According to the test control command, a normal engine response sample set and a normal platform response sample set are retrieved; features are extracted from the normal engine response sample set to obtain a normal engine response feature set, and statistical aggregation and benchmark calibration are performed on the normal engine response feature set to generate the engine response baseline feature matrix; features are extracted from the normal platform response sample set to obtain a normal platform response feature set, and statistical aggregation and benchmark calibration are performed on the normal platform response feature set to generate the platform response baseline feature matrix.

[0008] In a possible implementation, the candidate abnormal variable space is subjected to anomaly occurrence time-series tracking based on the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain, and the following processes are performed: the candidate abnormal variable space is temporally sorted to obtain the temporal adjacency relationship of variables; based on the temporal adjacency relationship of variables, the candidate abnormal variable space is mapped to the cross-domain variable propagation topology to obtain multiple abnormal variable propagation edges; the temporal continuity relationship, propagation direction relationship, and anomaly change direction relationship among the multiple abnormal variable propagation edges are identified; the multiple abnormal variable propagation edges are connected according to the temporal continuity relationship, the propagation direction relationship, and the anomaly change direction relationship to obtain the cross-domain anomaly propagation chain.

[0009] In a possible implementation, based on the cross-domain anomaly propagation chain, the matching degree of engine body fault propagation and the matching degree of test platform fault propagation are calculated, and the following processing is performed: feature extraction is performed on the cross-domain anomaly propagation chain to obtain a target propagation feature sequence; the engine body fault propagation template and the test platform fault propagation template corresponding to the test control command are retrieved; the target propagation feature sequence is matched with the engine body fault propagation template to obtain a first variable sequence matching value, a first propagation direction matching value, and a first propagation delay matching value; the first variable sequence matching value, the first propagation direction matching value, and the first propagation delay matching value are weighted and fused to generate the engine body fault propagation matching degree; the test platform fault propagation matching degree is calculated based on the target propagation feature sequence and the test platform fault propagation template.

[0010] In a possible implementation, a safety verification perturbation test is performed based on the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a bilateral test response sequence. The following processing is then performed: a difference calculation is performed on the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a propagation matching discrimination degree; if the propagation matching discrimination degree is less than or equal to a preset discrimination threshold, a disputed propagation edge is identified based on the cross-domain abnormal propagation chain; based on the disputed propagation edge, an initial verification test plan is generated, and the initial verification test plan is optimized for safety based on engine test safety constraints and platform test safety constraints to obtain an optimized verification test plan; response parameters of the engine body and the test platform under the optimized verification test plan are collected to obtain the bilateral test response sequence.

[0011] In a possible implementation, the fault source is located based on the test bilateral response sequence, and the fault source location result is obtained. The following processing is then performed: a response difference comparison is performed based on the test bilateral response sequence to generate a bilateral response difference feature set; based on the dispute propagation edge, a first bilateral expected response template under the engine body fault assumption and a second bilateral expected response template under the test platform fault assumption are fitted; the bilateral response difference feature set is registered and evaluated with the first bilateral expected response template to obtain engine body fault support; the bilateral response difference feature set is registered and evaluated with the second bilateral expected response template to obtain test platform fault support; the engine body fault support and the test platform fault support are compared to generate the fault source location result.

[0012] In a possible implementation, a cross-domain variable propagation topology is constructed between the engine body and the test platform, and the following processing is performed: information on the relationship between the engine body and the test platform is collected to obtain an ontology-platform relationship dataset; the cross-domain variable propagation topology is constructed based on the ontology-platform relationship dataset.

[0013] In a possible implementation, the test platform fault propagation matching degree is calculated based on the target propagation feature sequence and the test platform fault propagation template, and the following processing is performed: the target propagation feature sequence is matched with the test platform fault propagation template to obtain a second variable sequence matching value, a second propagation direction matching value, and a second propagation delay matching value; the second variable sequence matching value, the second propagation direction matching value, and the second propagation delay matching value are weighted and fused to generate the test platform fault propagation matching degree.

[0014] This application also provides a fault early warning system for an engine test platform based on time-series mining, including: a multi-source test data acquisition module for acquiring test control commands, engine body monitoring data, and test platform monitoring data during the test process; a response deviation detection and candidate anomaly space construction module for performing response deviation detection on the engine body monitoring data and the test platform monitoring data according to the test control commands, and establishing a candidate anomaly variable space; and a cross-domain propagation topology construction and time-series tracking module for constructing a cross-domain variable propagation topology between the engine body and the test platform, and performing response deviation detection and candidate anomaly space construction on the candidate anomaly variable space according to the cross-domain variable propagation topology. The system performs time-series tracking of anomalies in the quantum space to generate a cross-domain anomaly propagation chain; a fault propagation matching degree calculation module is used to calculate the fault propagation matching degree of the engine body and the test platform based on the cross-domain anomaly propagation chain; a safety verification disturbance test and bilateral response acquisition module is used to perform safety verification disturbance tests based on the fault propagation matching degree of the engine body and the test platform, and obtain the bilateral response sequence of the test; a fault source location and early warning command generation module is used to locate the fault source based on the bilateral response sequence of the test, obtain the fault source location result, and generate a fault early warning command based on the fault source location result.

[0015] The proposed method and system for fault early warning of engine test platforms based on time-series mining first acquires test control commands, engine body monitoring data, and test platform monitoring data during engine testing. Then, based on the test control commands, response deviation detection is performed on the engine body monitoring data and the test platform monitoring data to establish a candidate anomaly variable space. Next, a cross-domain variable propagation topology between the engine body and the test platform is constructed, and the anomaly occurrence time sequence is tracked in the candidate anomaly variable space based on the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain. Then, based on the cross-domain anomaly propagation chain, the fault propagation matching degree of the engine body and the test platform is calculated. Next, a safety verification disturbance test is performed based on the fault propagation matching degree of the engine body and the test platform to obtain a bilateral test response sequence. Finally, the fault source is located based on the bilateral test response sequence to obtain the fault source location result, and a fault early warning command is generated based on the fault source location result. Through the above process, the proposed method and system achieve the technical effect of accurately locating the fault source and improving the directionality and effectiveness of the early warning command under the condition of mutual coupling and interference between the engine body monitoring signal and the test platform monitoring signal. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the fault early warning method for an engine test platform based on time-series mining, provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of intermediate results for response deviation detection provided in an embodiment of this application.

[0019] Figure 3 This is a schematic diagram of the cross-domain anomaly propagation chain topology provided in an embodiment of this application.

[0020] Figure 4 A schematic diagram of the structure of the engine test platform fault early warning system based on time-series mining provided in this application embodiment.

[0021] Figure labeling: 10 Multi-source test data acquisition module, 20 Response deviation detection and candidate anomaly space construction module, 30 Cross-domain propagation topology construction and time series tracking module, 40 Fault propagation matching degree calculation module, 50 Security verification disturbance test and bilateral response acquisition module, 60 Fault source location and early warning command generation module. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0023] This application provides a fault early warning method for engine test platforms based on time-series mining, such as... Figure 1 As shown, the method includes: Step S100: Obtain test control commands, engine body monitoring data, and test platform monitoring data during the test process.

[0024] Specifically, three types of data are acquired in real time through the engine test bench's data acquisition system. Test control commands originate from digital command signals sent by the host computer to the engine control unit and test platform actuators, including speed setpoints, torque loading values, throttle opening commands, and dynamometer load commands, each with a timestamp. Engine body monitoring data is collected by sensors installed on the engine body, including but not limited to analog channels such as crankshaft speed, exhaust temperature of each cylinder, oil pressure, coolant temperature, knock sensor signal, and intake airflow. Each channel has a sampling frequency of at least 1kHz, and the data is converted from analog to digital to form a timestamped digital sequence. Test platform monitoring data is collected by sensors installed on platform equipment such as the test bench, dynamometer, intake air conditioning system, and exhaust back pressure regulating device. This includes measured dynamometer torque, measured dynamometer speed, bench vibration acceleration, measured intake air temperature, measured exhaust back pressure, and measured intercooler temperature, also collected synchronously at a sampling frequency of at least 1kHz. All data is time-stamped by a data acquisition card synchronized based on a precise time protocol and then stored in a real-time database in a time-aligned manner.

[0025] Step S200: Response deviation detection is performed on the engine body monitoring data and the test platform monitoring data according to the test control command, and a candidate abnormal variable space is established.

[0026] Specifically, using test control commands as a reference, the engine body monitoring data and test platform monitoring data are compared with their respective expected response benchmarks under normal conditions. This identifies monitoring variables whose response deviations exceed the normal fluctuation range and categorizes them into a candidate anomaly variable space. This step is used in the early stages of fault warning to quickly filter out a subset of potentially anomaly variables from a large number of real-time monitoring channels, narrowing the scope of subsequent analysis and avoiding indiscriminate exhaustive analysis of all monitoring variables, thereby improving the real-time performance and computational efficiency of fault diagnosis.

[0027] In one possible implementation, response deviation detection is performed on the engine body monitoring data and the test platform monitoring data according to the test control command, and a candidate abnormal variable space is established. Step S200 further includes step S210, which involves bilateral response baseline mining according to the test control command to establish an engine response baseline feature matrix and a platform response baseline feature matrix. Specifically, based on the current test control command type, historical response data of the engine and test platform under the same command and in normal operating conditions are retrieved from the historical database. Time-domain feature extraction, statistical aggregation, and benchmark calibration are performed on these normal response data respectively, ultimately forming independent response baseline feature matrices for the engine side and the platform side. This provides a standardized reference system for response deviation detection, enabling deviation detection to be based on the statistical regularity of the same model of equipment under the same command, rather than on a fixed threshold, thereby eliminating the interference caused by the differences in normal response waveforms under different operating conditions and commands on anomaly identification.

[0028] Step S220: Detect response deviations in the engine body monitoring data based on the engine response baseline feature matrix to obtain an engine response deviation sequence. Specifically, for the real-time data stream of the engine body monitoring data collected during the current test, a sliding time window method is used for response deviation detection. The sliding time window length is set to the time range of 300 sampling points after the current test control command is issued, and the window slides by one sampling point at a time. Within each time window, the real-time response data within the window is subjected to time-domain feature extraction as defined in step S212 to obtain the real-time values ​​of each feature in each channel under the current window. The real-time feature values ​​are compared one by one with the central reference value and normal fluctuation range of the corresponding feature in the corresponding channel of the engine response baseline feature matrix. If the real-time feature value exceeds the range of the central reference value plus or minus three standard deviations, the deviation value is calculated (the absolute value of the real-time feature value minus the central reference value divided by the standard deviation to obtain the standardized deviation factor), and this deviation is recorded as the response deviation value. The deviation values ​​of all channels within each time window constitute the deviation vector at that moment. The deviation vectors of all time windows are arranged in chronological order to form an engine response deviation sequence. Each element in the sequence consists of a timestamp, a monitoring channel identifier, and a standardized deviation factor.

[0029] Step S230: Detect response deviations in the test platform monitoring data based on the platform response baseline feature matrix to obtain a platform response deviation sequence. Specifically, for the real-time data stream of the test platform monitoring data collected during the current test, the same sliding time window method as in step S220 is used for processing. Within each time window, the same time-domain features as in step S213 are extracted from the real-time response data of each monitoring channel of the platform to obtain the real-time values ​​of each feature of each channel under the current window. The real-time feature values ​​are compared with the central reference value and normal fluctuation range of the corresponding feature of the corresponding channel in the platform response baseline feature matrix. If the values ​​exceed the range, the standardized deviation factor is calculated. The deviation values ​​of all platform channels within each time window constitute the platform deviation vector at that moment. The platform deviation vectors of all time windows are arranged in chronological order to form a platform response deviation sequence. Each element in the sequence consists of a timestamp, a platform monitoring channel identifier, and a standardized deviation factor.

[0030] Step S240: The engine response deviation sequence and the platform response deviation sequence are filtered according to the anomaly screening criteria to generate the candidate anomaly variable space. Specifically, the anomaly screening criteria are set as follows: the absolute value of the standardized deviation multiple is greater than 3, and the deviation state persists for more than 5 consecutive sampling points. The engine response deviation sequence is traversed to filter out all deviation events that meet the above criteria. Each deviation event is recorded with its start timestamp, duration, peak deviation multiple, and corresponding monitoring channel identifier. The same screening is performed on the platform response deviation sequence to obtain the set of deviation events on the platform side. The monitoring variables (including engine-side variables and platform-side variables) corresponding to all the screened deviation events are included in the candidate anomaly variable space. The candidate anomaly variable space is a set of variable identifiers, each of which is associated with its domain (engine domain or platform domain), the timestamp of the first occurrence of the anomaly, and the maximum deviation multiple. This step filters out noise interference by setting screening criteria to ensure that the anomaly variables entering the analysis are statistically significant.

[0031] For example, under the test condition of a certain model of six-cylinder diesel engine with a steady-state speed of 3000 r / min and a torque of 1000 N·m, the test control command is a sudden acceleration command, with the target throttle opening jumping from 30% to 70%. The data acquisition system records the response data at a sampling rate of 2 kHz. In step S210, normal response samples under this condition are retrieved from the historical database, totaling 50 sets of historical normal response records. In steps S220 and S230, time-domain features are extracted with a sliding window length of 300 sampling points (corresponding to 150 ms), and the normalization deviation multiple of each channel is obtained. Table 1 and Figure 2Intermediate results of deviation detection for four monitoring channels over 13 consecutive time windows are presented, with time windows numbered from 0 to 12, and each window interval being one sampling period (0.5ms).

[0032] Table 1

[0033] In step S240, the anomaly screening condition is that the absolute value of the deviation multiple is greater than 3 and lasts for more than 5 sampling points. The engine speed deviation multiple exceeds the threshold in windows 4 to 6 but only lasts for 3 windows (windows 4, 5, and 6), which does not meet the requirement of lasting 5 sampling points, therefore it is not included in the candidate anomaly variable space. The engine exhaust temperature deviation multiple continuously exceeds the threshold in windows 7 to 12, and the number of consecutive windows exceeds 5 sampling points, therefore the exhaust temperature variable is included in the candidate anomaly variable space, and its first anomaly time is recorded as the start time of window 7. The dynamometer torque deviation multiple continuously exceeds the threshold in windows 7 to 12, and the number of consecutive windows exceeds 5 sampling points, therefore it is included in the candidate anomaly variable space, and the first anomaly time is recorded as the start time of window 7. The bench vibration deviation multiple does not exceed the threshold throughout the entire process and is not included. The output of this example is a candidate anomaly variable space containing two variables: exhaust temperature and dynamometer torque, belonging to the engine domain and platform domain respectively.

[0034] In one possible implementation, bilateral response baseline mining is performed according to the test control command to establish an engine response baseline feature matrix and a platform response baseline feature matrix. Step S210 further includes step S211, which retrieves the engine normal response sample set and the platform normal response sample set according to the test control command. Specifically, historical test data of this engine model under calibration conditions is retrieved from the historical database, and test records marked as normal operating conditions are filtered. Based on the type and numerical range of the current test control command, historical response data segments under the same or similar commands are determined. Specifically, using the category code and time series length of the current control command as search conditions, the response waveforms of each monitoring channel of the engine body and the response waveforms of each monitoring channel of the test platform under the corresponding command are retrieved from the historical normal test records, forming the engine normal response sample set and the platform normal response sample set, respectively. Each sample contains the numerical sequence of each monitoring channel within a time window from 50 sampling points before the command is issued to 500 sampling points after the command is issued. For example, if the current command is a speed step command and the target speed is 3000 r / min, then retrieve the normal step response records under the same target speed in the historical data. Each record contains the complete response waveform before and after the time the command was issued.

[0035] Step S212: Feature extraction is performed based on the engine normal response sample set to obtain an engine normal response feature set. Statistical aggregation and benchmark calibration are then performed on the engine normal response feature set to generate the engine response baseline feature matrix. Specifically, a time-domain feature extraction method is used to extract features from each sample in the engine normal response sample set. For the response curve of each monitoring channel under a step command, the following time-domain features are extracted: response delay time (the time difference from the command issuance to the first deviation of the response value from the initial steady-state value threshold of 2%), rise time (the time difference from 10% to 90% of the response value reaching the final value), peak overshoot (the percentage of the difference between the maximum and final response values), steady-state error (the difference between the mean and the expected command value after the response stabilizes), the maximum value of the first derivative of the response curve, and the absolute value of the integral. After extraction, the same feature for the same channel of all samples constitutes a feature value sequence. Statistical aggregation is performed on this feature value sequence: the arithmetic mean, standard deviation, and median are calculated. The mean is used as the central benchmark value for this feature, and three times the standard deviation is used as the upper limit of normal fluctuation for this feature. The central baseline values ​​of all features across all channels are arranged according to a two-dimensional index of channel number and feature type, forming an engine response baseline feature matrix. Each element in the matrix corresponds to the baseline value and normal fluctuation range of a feature type for a monitoring channel. This step establishes a normal response baseline through statistical aggregation, providing a quantitative reference standard for deviation detection.

[0036] Step S213: Feature extraction is performed based on the platform's normal response sample set to obtain a platform normal response feature set. Statistical aggregation and benchmark calibration are then performed on this feature set to generate the platform response baseline feature matrix. Specifically, the same time-domain feature extraction operation as in step S212 is performed on each sample in the test platform's normal response sample set. The extraction targets are the response curves of each monitoring channel of the test platform. The extracted feature types are consistent with those in step S212, including response delay time, rise time, peak overshoot, steady-state error, maximum value of the first derivative, and absolute value of the integral. Statistical aggregation is performed on each feature of each channel after extraction: the arithmetic mean, standard deviation, and median are calculated. The mean is used as the central benchmark value, and three times the standard deviation is used as the upper limit of normal fluctuation. The central benchmark values ​​and normal fluctuation ranges of all features of all platform monitoring channels are arranged according to a two-dimensional index of channel number and feature type to form the platform response baseline feature matrix. This matrix has the same feature type index structure as the engine response baseline feature matrix, but corresponds to a different set of monitoring channels.

[0037] Step S300: Construct a cross-domain variable propagation topology between the engine body and the test platform, and perform anomaly occurrence timing tracking on the candidate abnormal variable space based on the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain.

[0038] Specifically, a topology graph containing engine domain nodes and platform domain nodes and their interconnections is constructed using bench design documents and physical connection layout information. Anomalies in the candidate anomaly variable space are mapped onto this topology graph according to the chronological order of their occurrence for path tracing. This identifies which anomalies are correlated in time and space, and these correlations are linked into one or more anomaly propagation chains spanning the engine and platform domains. This step organizes previously isolated single-point anomaly signals into structured information with temporal logic and propagation direction, providing quantifiable propagation patterns to determine whether the fault originates in the engine itself or the test platform.

[0039] In one possible implementation, a cross-domain variable propagation topology is constructed between the engine body and the test platform. Step S300 further includes step S310, which involves collecting relationship information between the engine body and the test platform to obtain an engine body-platform relationship dataset. Specifically, coupling relationship information between the engine body and various devices on the test platform is collected from the test bench design document and physical connection layout, including mechanical connection relationships, thermal coupling relationships, and electrical connection relationships. Mechanical connection relationships include the coupling connection between the engine output shaft and the dynamometer input shaft, the fixed connection between the engine support base and the test bench foundation, and the connections between various pipe flanges. Each connection relationship records the names of the components connected at both ends and the connection type. Thermal coupling relationships include the heat conduction path between the engine exhaust manifold and the exhaust pipe, the fluid path between the engine coolant outlet and the heat exchanger, and the heat exchange path between the intercooler and the intake pipe. Each thermal path records the direction of heat transfer and the intermediate components along the path. Electrical connection relationships include the communication bus connection between the engine control unit and the dynamometer controller, and the signal line connection between the sensor signal conditioner and the data acquisition card. Each electrical connection records the signal source end, the receiver end, and the signal type. All relational data is organized in the form of triples (first component, relation type, second component) to form the ontology platform relational dataset.

[0040] Step S320: Construct the cross-domain variable propagation topology based on the ontology platform relationship dataset. Specifically, each component in the ontology platform relationship dataset is used as a topology node, and the connections between components are used as topology edges to construct an undirected topology graph. For each monitored variable, it is mapped to its corresponding component node; that is, the engine body monitoring variable is mapped to the engine body node, the dynamometer torque measurement value is mapped to the dynamometer node, the vibration acceleration is mapped to the test bench foundation node, and the temperature sensor value is mapped to the corresponding pipeline node. All nodes in the topology graph are divided into an engine domain node set and a platform domain node set, and the two domains are interconnected through mechanical connection edges, thermal path edges, and electrical connection edges. For each connection edge, a base value for propagation delay is assigned according to the physical characteristics of its connection type: the base value for propagation delay of mechanical connection edges is 10ms, the base value for propagation delay of thermal path edges is 500ms, and the base value for propagation delay of electrical connection edges is 1ms. The topology graph with nodes and edges constructed in this way is the cross-domain variable propagation topology, used to track the propagation path of anomalies between different domains.

[0041] In one possible implementation, the candidate abnormal variable space is subjected to anomaly occurrence timing tracking based on the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain. Step S300 further includes step S330, which involves temporally sorting the candidate abnormal variable space to obtain the temporal adjacency relationship of variables. Specifically, the timestamp of the first occurrence of each abnormal variable is extracted from the candidate abnormal variable space, and all abnormal variables are arranged in ascending order of these timestamps to form a temporal sorting list. For two adjacent abnormal variables in the temporal sorting list, their timestamp difference is calculated. If the difference is less than a preset temporal adjacency threshold (e.g., set to 200ms), it is determined that there is a temporal adjacency relationship between the two variables, and this adjacency relationship is recorded as a temporal adjacency pair, including the preceding variable identifier, the following variable identifier, and the time interval value. The entire temporal sorting list is traversed to obtain a set of all temporal adjacency relationships that meet the conditions.

[0042] Step S340: Based on the temporal adjacency relationship of the variables, the candidate abnormal variable space is mapped to the cross-domain variable propagation topology to obtain multiple abnormal variable propagation edges. Specifically, for each temporal adjacency pair obtained in step S330, the shortest path between the topological nodes corresponding to the two abnormal variables is found in the cross-domain variable propagation topology. If there is at least one connected path between the two nodes, the temporal adjacency pair is mapped as an abnormal variable propagation edge. The abnormal variable propagation edge records the propagation start variable, the propagation end variable, the shortest path length between the two in the topology (in terms of the number of edges), and the actual time interval. Temporal adjacency pairs that cannot find a connected path in the topology are discarded and not used as abnormal variable propagation edges. After traversing all temporal adjacency pairs, a set of multiple abnormal variable propagation edges is obtained.

[0043] Step S350: Identify the temporal continuity, propagation direction, and abnormal change direction relationships among the multiple abnormal variable propagation edges. Specifically, perform pairwise comparisons on the set of abnormal variable propagation edges obtained in step S340. For two abnormal variable propagation edges, if the propagation termination variable of the first edge and the propagation initiation variable of the second edge share the same variable identifier, then a temporal continuity relationship exists between these two edges. For each abnormal variable propagation edge, determine the propagation direction based on the order of the first abnormal timestamp of its propagation initiation variable and the first abnormal timestamp of its propagation termination variable. If the initiation variable's abnormality occurs first, the propagation direction is from the initiation variable to the termination variable. For each abnormal variable propagation edge, compare the deviation change direction at the time the initiation variable's abnormality occurs with the deviation change direction at the time the termination variable's abnormality occurs. If both standardized deviation factors are positive or both are negative, the abnormal change direction relationship is in the same direction; otherwise, it is in opposite directions. Perform the above identification process for all edges one by one, recording the propagation direction marker and change direction marker for each edge.

[0044] Step S360: Connect the multiple abnormal variable propagation edges according to the temporal continuity relationship, the propagation direction relationship, and the abnormal change direction relationship to obtain the cross-domain abnormal propagation chain. Specifically, using the abnormal variable propagation edges in step S340 as basic units, these edges are connected end-to-end according to the temporal continuity relationship. That is, starting from the edge where the earliest abnormality occurred, the next edge with the same variable at the end of the current edge is continuously searched. If multiple candidate edges exist, edges with the same propagation direction label and the same abnormal change direction relationship are preferentially selected for connection. Each time an edge is connected, the current propagation chain is extended by one edge unit. When it is impossible to continue connecting, the current edge sequence constitutes a complete cross-domain abnormal propagation chain. The cross-domain abnormal propagation chain contains multiple abnormal variable nodes arranged in chronological order and directed edges connecting these nodes. Each directed edge is labeled with the propagation direction, time interval, and abnormal change direction. If the propagation chain contains both engine domain variable nodes and platform domain variable nodes, the propagation chain is considered a cross-domain propagation chain, and the final output is one or more cross-domain abnormal propagation chains.

[0045] For example, such as Figure 3As shown, the candidate abnormal variable space is set to include the following abnormal variables sorted by time: the first abnormal variable is the engine exhaust temperature (engine domain), with an abnormal occurrence timestamp of 1050ms; the second abnormal variable is the measured torque value of the dynamometer (platform domain), with an abnormal occurrence timestamp of 1080ms; the third abnormal variable is the engine coolant outlet temperature (engine domain), with an abnormal occurrence timestamp of 1120ms; and the fourth abnormal variable is the vibration acceleration of the test bench foundation (platform domain), with an abnormal occurrence timestamp of 1150ms. In step S330, the time sequence sorting result is as described above. Calculate the difference between adjacent timestamps: 2nd time stamp minus 1st time stamp is 30ms, 3rd time stamp minus 2nd time stamp is 40ms, 4th time stamp minus 3rd time stamp is 30ms, all less than the preset temporal adjacency threshold of 200ms. Therefore, three temporal adjacency relationships are obtained: (exhaust temperature, dynamometer torque, 30ms), (dynamometer torque, coolant temperature, 40ms), (coolant temperature, test bench vibration, 30ms). In step S340, the cross-domain variable propagation topology includes the following nodes and paths: the engine node is connected to the exhaust pipe node, the exhaust pipe node is connected to the dynamometer torque sensor node via a heat conduction edge, the dynamometer torque sensor node is connected to the test bench foundation node via a mechanical connection edge; simultaneously, the engine node is connected to the coolant temperature sensor node via the coolant pipe node, the coolant temperature sensor node is connected to the data acquisition node via an electrical connection edge, and the data acquisition node is connected to the test bench vibration sensor node via an electrical connection edge. All three temporally adjacent pairs have a retrievable connected path in the topology, thus generating three anomalous variable propagation edges: Edge 1 from exhaust temperature to dynamometer torque, path length 1, time interval 30ms; Edge 2 from dynamometer torque to coolant temperature, path length 4, time interval 40ms; Edge 3 from coolant temperature to bench vibration, path length 2, time interval 30ms. In step S350, temporal continuity is identified: the terminating variable of Edge 1 is dynamometer torque, and the starting variable of Edge 2 is dynamometer torque; since they are the same, Edge 1 and Edge 2 have a temporal continuity relationship; the terminating variable of Edge 2 is coolant temperature, and the starting variable of Edge 3 is coolant temperature, so Edge 2 and Edge 3 have a temporal continuity relationship. Propagation direction relationship: Edge 1 starts from an engine domain variable and ends from a platform domain variable, marked as engine domain to platform domain direction; Edge 2 starts from a platform domain variable and ends from an engine domain variable, marked as platform domain to engine domain direction; Edge 3 starts from an engine domain variable and ends from a platform domain variable, marked as engine domain to platform domain direction. Abnormal change direction relationship: The standardized deviation multiples of the variables at the starting and ending points of the three sides are all positive, and all belong to the same direction of change relationship.In step S360, edges 1, 2, and 3 are sequentially connected in a temporal sequence to generate a complete cross-domain anomaly propagation chain: exhaust temperature → dynamometer torque (30ms, engine domain to platform domain) → coolant temperature (40ms, platform domain to engine domain) → bench vibration (30ms, engine domain to platform domain). This propagation chain simultaneously covers variable nodes in both the engine domain and the platform domain, with the anomaly propagation direction alternating between the two domains.

[0046] Step S400: Calculate the engine body fault propagation matching degree and the test platform fault propagation matching degree based on the cross-domain anomaly propagation chain.

[0047] Specifically, three types of features—variable order, propagation direction, and propagation delay—are extracted from the cross-domain anomaly propagation chain to form a target propagation feature sequence. This sequence is then matched and compared with pre-stored engine body fault propagation templates and test platform fault propagation templates. Matching values ​​are calculated for three dimensions: variable order consistency, propagation direction consistency, and propagation delay consistency. These three dimensions are then weighted and fused to obtain two quantitative indicators: engine body fault propagation matching degree and test platform fault propagation matching degree. This step transforms the qualitative structure of the anomaly propagation chain into two comparable quantitative scores, serving as a preliminary basis for judging the fault source tendency. When the matching degree of one side is significantly higher than that of the other side, it can be preliminarily determined that the fault is biased towards that side.

[0048] In one possible implementation, based on the cross-domain anomaly propagation chain, the engine body fault propagation matching degree and the test platform fault propagation matching degree are calculated. Step S400 further includes step S410, which extracts features from the cross-domain anomaly propagation chain to obtain a target propagation feature sequence. Specifically, three types of propagation features are extracted from each cross-domain anomaly propagation chain obtained in step S360: variable order features, propagation direction features, and propagation delay features. The variable order feature is a sequence of variable identifiers arranged chronologically from the anomaly variable nodes in the propagation chain; the propagation direction feature is a sequence of direction markers for each directed edge in the propagation chain, where the direction marker value is from the engine domain to the platform domain or from the platform domain to the engine domain; and the propagation delay feature is a sequence of time interval values ​​corresponding to each directed edge in the propagation chain. These three types of features are aligned and arranged according to edge order to form the target propagation feature sequence. The i-th element in the sequence contains the i-th anomaly variable identifier, the direction marker of the i-th edge, and the time interval value of the i-th edge. For a propagation chain containing k anomaly variable nodes, the target propagation feature sequence contains feature information from (k-1) edges.

[0049] Step S420: Retrieve the engine body fault propagation template and the test platform fault propagation template corresponding to the test control command. Specifically, retrieve the two corresponding fault propagation templates from the fault template library according to the type and value range of the current test control command. The fault template library pre-stores engine body fault propagation templates and test platform fault propagation templates for this engine model under various typical control commands. The engine body fault propagation template records the expected sequence, propagation direction, and time interval range of each monitored variable when typical engine body faults occur, such as single-cylinder misfire, turbocharger overspeed, or crankshaft bearing wear. The test platform fault propagation template records the expected sequence, propagation direction, and time interval range of each monitored variable when typical test platform faults occur, such as dynamometer torque fluctuation, bench resonance, or intake air conditioning failure. Each template is stored in the form of a propagation feature sequence, with the same data structure as the target propagation feature sequence in step S410.

[0050] Step S430: Match the target propagation feature sequence with the engine body fault propagation template to obtain a first variable order matching value, a first propagation direction matching value, and a first propagation delay matching value. Specifically, the longest common subsequence algorithm is used to calculate the degree of variable order matching between the target propagation feature sequence and the engine body fault propagation template. That is, taking the variable identifier sequence of the target propagation feature sequence as the first sequence and the variable identifier sequence of the engine body fault propagation template as the second sequence, the length of the longest common subsequence of the two sequences is calculated. This length is divided by the total length of the second sequence to obtain the first variable order matching value, which ranges from 0% to 100%. For propagation direction matching, the direction marker of each edge in the target propagation feature sequence is compared with the direction marker of the corresponding sequential position in the engine body fault propagation template. The number of edges with the same direction marker is counted, and this number is divided by the total number of direction markers in the template to obtain the first propagation direction matching value. For propagation delay matching, the time interval value of each edge in the target propagation feature sequence is compared with the time interval reference value of the corresponding sequential position in the engine body fault propagation template. Each time interval reference value in the template includes an allowable deviation range, defined as ±20% of the reference value. If the target time interval value falls within this allowable deviation range, the edge delay is successfully matched. The number of successfully matched edges is counted and divided by the total number of time interval reference values ​​in the template to obtain the first propagation delay matching value.

[0051] Step S440: The first variable sequence matching value, the first propagation direction matching value, and the first propagation delay matching value are weighted and fused to generate the engine body fault propagation matching degree. Specifically, the weight of the variable sequence matching value is set to 40%, the weight of the propagation direction matching value to 30%, and the weight of the propagation delay matching value to 30%. Engine body fault propagation matching degree = first variable sequence matching value × 40% + first propagation direction matching value × 30% + first propagation delay matching value × 30%, with the result expressed as a percentage.

[0052] Step S450: Calculate the test platform fault propagation matching degree based on the target propagation feature sequence and the test platform fault propagation template. Specifically, perform three-dimensional matching calculations between the target propagation feature sequence extracted in step S410 and the test platform fault propagation template retrieved from the fault template library: variable order matching value, propagation direction matching value, and propagation delay matching value. Then, weightedly fuse these three matching values ​​to obtain the test platform fault propagation matching degree.

[0053] In one possible implementation, based on the target propagation feature sequence and the test platform fault propagation template, the test platform fault propagation matching degree is calculated. Step S450 further includes step S451, matching the target propagation feature sequence with the test platform fault propagation template to obtain a second variable order matching value, a second propagation direction matching value, and a second propagation delay matching value. Specifically, using the same matching calculation method as in step S430, the variable identifier sequence of the target propagation feature sequence is matched with the variable identifier sequence of the test platform fault propagation template using the longest common subsequence matching to calculate the second variable order matching value. The direction marker sequence of the target propagation feature sequence is compared bit by bit with the direction marker sequence of the test platform fault propagation template, and the consistency ratio is calculated to obtain the second propagation direction matching value. The time interval value of the target propagation feature sequence is compared bit by bit with the time interval reference value of the test platform fault propagation template (allowable deviation range is ±20%), and the proportion of edges falling within the allowable range is calculated to obtain the second propagation delay matching value.

[0054] Step S452 involves weighted fusion of the second variable sequence matching value, the second propagation direction matching value, and the second propagation delay matching value to generate the test platform fault propagation matching degree. Specifically, the same weight configuration as in step S440 is used, i.e., the variable sequence matching value has a weight of 40%, the propagation direction matching value has a weight of 30%, and the propagation delay matching value has a weight of 30%. The test platform fault propagation matching degree = second variable sequence matching value × 40% + second propagation direction matching value × 30% + second propagation delay matching value × 30%, with the result expressed as a percentage.

[0055] For example, let's continue with the cross-domain anomaly propagation chain in step S360. This propagation chain contains 4 anomaly variable nodes and 3 directed edges. In step S410, the target propagation feature sequence is extracted as follows: edge 1 (exhaust temperature → dynamometer torque, engine domain to platform domain direction, delay 30ms), edge 2 (dynamometer torque → coolant temperature, platform domain to engine domain direction, delay 40ms), and edge 3 (coolant temperature → test bench vibration, engine domain to platform domain direction, delay 30ms). In step S420, the engine body fault propagation template under the current throttle step command is retrieved. The variable sequence of this template is (exhaust temperature → coolant temperature → engine oil temperature), the direction sequence is (engine domain to engine domain, engine domain to engine domain), and the delay reference value sequence is (50ms, 60ms). The test platform fault propagation template is retrieved. The variable sequence of this template is (dynamometer torque → bench vibration → intake air temperature), the direction sequence is (platform domain to platform domain, platform domain to platform domain), and the time delay reference value sequence is (20ms, 25ms). In step S430, the target feature sequence variable identifiers are (exhaust temperature, dynamometer torque, coolant temperature, bench vibration), and the engine body fault template variable identifiers are (exhaust temperature, coolant temperature, engine oil temperature). The longest common subsequence of the two is (exhaust temperature, coolant temperature), with a sequence length of 2. The total length of the template variables is 3, therefore the first variable sequence matching value is 66.67%. During the propagation direction matching comparison, the target direction sequence is (engine domain to platform domain, platform domain to engine domain, engine domain to platform domain). The engine template only contains two edge direction sequences (engine domain to engine domain, engine domain to engine domain). The first two edges are selected for comparison bit by bit. The direction labels of the two edges are inconsistent, the number of successfully matched edges is 0, and the first propagation direction matching value is 0%. During time delay matching, the first two time delay values ​​(30ms and 40ms) of the target sequence are compared with the template time delay reference value. The allowable deviation range for time delay is ±20% of the reference value. The allowable interval for 50ms is 40ms~60ms, and 30ms does not fall into this interval. The allowable interval for 60ms is 48ms~72ms, and 40ms does not fall into this interval. Both edge time delays fail to match, the number of successfully matched edges is 0, and the first propagation time delay matching value is 0%. In step S451, the target feature sequence is compared with the test platform fault propagation template: the longest common subsequence of the variable sequence is (dynamometer torque, test bench vibration), with a length of 2. The total length of the template variables is 3, and the second variable sequence matching value is 66.67%. During direction sequence matching, the first two edge directions of the target are (engine domain to platform domain, platform domain to engine domain), and the platform template direction sequence is (platform domain to platform domain, platform domain to platform domain). There are no matching edges, and the second propagation direction matching value is 0%.During the delay matching comparison, the first two delays of the target (30ms, 40ms) are compared one by one with the template delay reference values ​​(20ms, 25ms, allowable deviation ±20%, corresponding to intervals of 16ms~24ms and 20ms~30ms). Since neither 30ms nor 40ms falls within the corresponding interval, the number of successful delay matching edges is 0, and the second propagation delay matching value is 0%. In step S440, the engine body fault propagation matching degree is calculated according to the weighted formula: 66.67%×40%+0%×30%+0%×30%=26.67%. In step S452, the test platform fault propagation matching degree is calculated: 66.67%×40%+0%×30%+0%×30%=26.67%.

[0056] Step S500: Perform a safety verification disturbance test based on the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain the test bilateral response sequence.

[0057] Specifically, the difference between two matching degrees is calculated as the discrimination index. If the discrimination index is lower than a preset threshold, it indicates that the current passive monitoring data is insufficient to distinguish the source of the fault. In this case, the disputed propagation edge is identified from the cross-domain anomaly propagation chain, and a verification test plan is generated based on the variable types involved in the disputed propagation edge. The plan is then optimized for amplitude limiting and waveform smoothing according to the safety constraints of the engine and test platform. Finally, the optimized excitation signal is sent out, and response data from the engine side and the platform side are collected simultaneously to form a bilateral test response sequence. This step breaks the deadlock of matching degree under passive monitoring by actively perturbing the system to elicit differentiated response characteristics under two fault assumptions, thereby providing more discriminative data support for the final fault source location.

[0058] In one possible implementation, a safety verification disturbance test is performed based on the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a bilateral test response sequence. Step S500 further includes step S510, calculating the difference between the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a propagation matching discrimination degree. Specifically, the propagation matching discrimination degree is obtained by subtracting the test platform fault propagation matching degree obtained in step S452 from the engine body fault propagation matching degree obtained in step S440, and taking the absolute value. This is expressed as a percentage.

[0059] Step S520: If the propagation matching discrimination is less than or equal to a preset discrimination threshold, identify disputed propagation edges according to the cross-domain anomaly propagation chain. Specifically, the preset discrimination threshold is set to 15%. If the propagation matching discrimination is less than or equal to 15%, it indicates that the matching degree of the two fault hypotheses is close, and the source of the fault cannot be reliably distinguished. At this time, it is necessary to identify disputes on the propagation edges in the cross-domain anomaly propagation chain. The method for identifying disputed propagation edges is as follows: For each directed edge in the cross-domain anomaly propagation chain, calculate the time delay matching score of the edge in the engine body fault propagation template and the time delay matching score in the test platform fault propagation template. If the difference between the two scores is less than or equal to 10%, the edge is marked as a disputed propagation edge. All edges marked as disputed propagation edges are combined into a disputed propagation edge set.

[0060] Step S530: Based on the dispute propagation edges, an initial verification test plan is generated, and the initial verification test plan is optimized for safety according to engine test safety constraints and platform test safety constraints to obtain an optimized verification test plan. Specifically, based on the variable types involved in the dispute propagation edge set, corresponding verification test excitation signals are generated. If the variables involved in the dispute propagation edges include engine speed and dynamometer torque, the initial verification test plan is to superimpose a speed step disturbance signal with an amplitude of 5% of the current speed and a torque sinusoidal disturbance signal with an amplitude of 5% of the current torque at the current steady-state operating point of the engine, with a disturbance duration of 2 seconds. If the variables involved in the dispute propagation edges are mainly temperature-related variables, such as exhaust temperature and coolant temperature, the initial verification test plan is to apply a temperature step disturbance signal with an amplitude of ±2℃ at the air intake, with a disturbance duration of 5 seconds. After generating the initial verification test plan, safety optimization is performed based on engine test safety constraints and platform test safety constraints. Engine testing safety constraints include engine speed not exceeding the maximum permissible speed for the engine model (e.g., 6000 r / min), exhaust temperature not exceeding the maximum permissible exhaust temperature (e.g., 850℃), and oil pressure not falling below the minimum permissible pressure (e.g., 200 kPa). Platform testing safety constraints include dynamometer torque not exceeding the dynamometer's rated torque (e.g., 3000 N·m), bench vibration acceleration not exceeding the safety limit (e.g., 5g), and intake air conditioning outlet temperature change rate not exceeding 2℃ per second. The safety optimization process involves comparing the disturbance amplitude in the initial verification test plan with the margins of the aforementioned safety constraints one by one. If the disturbance amplitude exceeds the permissible margin of any safety constraint, the disturbance amplitude is proportionally reduced until all safety constraints are met. The disturbance waveform is smoothed using a second-order low-pass filter to eliminate step abrupt changes, with the filter cutoff frequency set to 10 Hz. The test plan after safety optimization is the verification test optimization plan, which includes the excitation signal type, amplitude, waveform, and time parameters after safety tuning.

[0061] Step S540: Collect response parameters of the engine body and test platform under the verification test optimization scheme to obtain the bilateral test response sequence. Specifically, according to the time parameters in the verification test optimization scheme, excitation signals are sent to the engine control unit and dynamometer controller, synchronously triggering the data acquisition system to record all data from 100 sampling points before the excitation starts to 200 sampling points after the excitation ends. The collected response parameters include real-time values ​​of the monitoring channels on the engine body side and the monitoring channels on the test platform side. The response values ​​on the engine body side are arranged in order of sampling time to form an engine-side response subsequence; the response values ​​on the test platform side are arranged in order of sampling time to form a platform-side response subsequence. After aligning the engine-side response subsequence and the platform-side response subsequence with the same timestamp, they are merged into a bilateral test response sequence. Each element in the sequence consists of a timestamp, a value vector of each channel on the engine side, and a value vector of each channel on the platform side.

[0062] Step S600: Locate the fault source based on the test bilateral response sequence, obtain the fault source location result, and generate a fault warning instruction based on the fault source location result.

[0063] Specifically, the bilateral response sequences collected under disturbance testing are compared with the expected responses under normal conditions to extract a set of bilateral response difference features. This set of features is then registered and evaluated with expected response templates under both the engine body fault assumption and the test platform fault assumption to obtain fault support under the two assumptions. Finally, the magnitudes of the two support values ​​are compared, and the one with higher support is taken as the fault source location result. Based on this result, a fault warning instruction containing a fault source identifier and specific inspection items is generated. This step integrates all the discrimination information accumulated in the previous steps into a clear fault source determination conclusion and outputs it in the form of an engineering-usable warning instruction. This allows test operators to quickly locate the troubleshooting direction based on the instruction, reducing the time consumption and safety risks caused by blind troubleshooting.

[0064] In one possible implementation, fault source localization is performed based on the test bilateral response sequence to obtain the fault source localization result. Step S600 further includes step S610, which compares response differences based on the test bilateral response sequence to generate a bilateral response difference feature set. Specifically, the engine-side response subsequence in the test bilateral response sequence is subtracted point by point from the average response curve of the engine normal response sample set retrieved in step S210 under the same excitation condition to obtain an engine-side response difference sequence, where each time point corresponds to a difference vector. Difference features are extracted from the engine-side response difference sequence: the peak value of the absolute difference, the area of ​​the integral of the absolute difference, the maximum rising slope of the difference sequence, and the maximum falling slope of the difference sequence are calculated. Using the same method, the platform-side response subsequence in the test bilateral response sequence is subtracted point by point from the average response curve of the platform normal response sample set under the same excitation condition to obtain a platform-side response difference sequence, and the same defined peak value, area of ​​the integral of the difference, rising slope, and falling slope are extracted. All extracted difference features are combined into a bilateral response difference feature set according to their domain (engine domain or platform domain) and feature type.

[0065] Step S620: Based on the dispute propagation edge, fit a first bilateral expected response template under the engine body fault assumption and a second bilateral expected response template under the test platform fault assumption. Specifically, based on the fault propagation path indicated by the dispute propagation edge, retrieve the transfer function model corresponding to the propagation path from the fault effect database. For the engine body fault assumption, retrieve the engine body fault transfer function, which describes the response propagation relationship from the engine body fault source to each monitoring point. This transfer function is stored in the form of a transfer function matrix, where the matrix elements are the gain coefficient and time constant of each channel relative to the fault source excitation. Using the initial variable of the dispute propagation edge as the fault injection point, apply the transfer function matrix to the input excitation signal of the verification test optimization scheme to calculate the expected response waveforms of each channel on the engine side and each channel on the platform side. Combine these two to form the first bilateral expected response template, which includes the expected waveforms on the engine side and the expected waveforms on the platform side. For the test platform fault assumption, retrieve the test platform fault transfer function, which describes the response propagation relationship from the test platform fault source to each monitoring point. The starting variable of the dispute propagation edge is used as the fault injection point (but the fault source is assumed to be the platform side). The transfer function matrix is ​​applied to the input excitation signal of the same verification test optimization scheme to calculate the expected response waveform on the engine side and the expected response waveform on the platform side. The two are combined into a second bilateral expected response template.

[0066] Step S630: The bilateral response difference feature set is registered and evaluated with the first bilateral expected response template to obtain the engine body fault support. Specifically, the engine-side difference peak value, difference integral area, rising edge slope, and falling edge slope in the bilateral response difference feature set are compared with the expected difference features of the corresponding channels in the first bilateral expected response template. For each difference feature, the ratio of the real-time difference feature value to the expected difference feature value is calculated. If the ratio is between 0.5 and 2.0, the feature is successfully registered. The proportion of successfully registered features to the total number of features is counted to obtain the registration success rate. This registration success rate is multiplied by the average matching degree of the edges not marked as dispute propagation edges in step S520 (this average value is obtained by averaging the matching values ​​of the non-disputed edges in step S430) to obtain the engine body fault support, expressed as a percentage.

[0067] Step S640: The bilateral response difference feature set is registered and evaluated with the second bilateral expected response template to obtain the test platform fault support. Specifically, using the same registration evaluation method as in step S630, each difference feature in the bilateral response difference feature set is compared one by one with the expected difference features of the corresponding channel in the second bilateral expected response template. The percentage of successfully registered features is counted to obtain the registration success rate. This registration success rate is multiplied by the average matching degree of the edges not marked as dispute propagation edges in step S520 (this average value is obtained by averaging the matching values ​​of the non-disputed edges in step S451) to obtain the test platform fault support, expressed as a percentage.

[0068] Step S650: Compare the fault support of the engine body and the fault support of the test platform to generate the fault source localization result. Specifically, compare the magnitudes of the fault support of the engine body and the fault support of the test platform. If the fault support of the engine body is greater than that of the test platform, the fault source is determined to be the engine body; if the fault support of the test platform is greater than that of the engine body, the fault source is determined to be the test platform; if the absolute value of the difference between the two is less than 5%, it is determined that the fault source cannot be uniquely determined, and an uncertainty flag is output. Based on the fault location results, corresponding fault warning instructions are generated: If the fault is determined to be an engine body fault, an engine body fault warning instruction is generated, which includes the fault source identifier as engine and a list of inspection items (including ignition system, fuel injection system, turbocharging system, and crankshaft connecting rod mechanism); if the fault is determined to be a test platform fault, a test platform fault warning instruction is generated, which includes the fault source identifier as platform and a list of inspection items (including dynamometer coupling, test bench foundation fasteners, intake air conditioning, and exhaust back pressure regulating valve); if the fault cannot be determined, a dual-end troubleshooting warning instruction is generated, which includes a prompt that both engine and platform-related items need to be checked simultaneously.

[0069] This application's embodiments solve the technical problem of insufficient warning directionality in existing engine test fault warnings due to unclear monitoring signal sources. This is achieved by acquiring three types of raw data during the testing process, detecting and filtering abnormal variables through response deviation detection of two types of monitoring data, constructing a cross-domain topology and generating a propagation chain through time-series tracking of abnormal variables, calculating the propagation matching degree under two fault assumptions, performing safety disturbance tests and collecting bilateral responses when the matching degree is close, and finally locating the fault source and outputting warning commands based on the response data. This achieves the technical effect of accurately locating the fault source and improving the directionality and effectiveness of warning commands even under conditions of mutual coupling and interference between engine body monitoring signals and test platform monitoring signals.

[0070] In the above text, refer to Figures 1-3 This paper describes in detail a fault early warning method for an engine test platform based on time-series mining according to an embodiment of the present invention. Next, we will refer to... Figure 4 A fault early warning system for an engine test platform based on time-series mining, according to an embodiment of the present invention, is described.

[0071] The engine test platform fault early warning system based on time-series mining according to embodiments of the present invention addresses the technical problem of insufficient early warning directionality caused by unclear monitoring signal sources in existing engine test fault early warning systems. It achieves the technical effect of accurately locating the fault source and improving the directionality and effectiveness of early warning commands even under conditions of mutual coupling and interference between engine body monitoring signals and test platform monitoring signals. The engine test platform fault early warning system based on time-series mining includes: a multi-source test data acquisition module 10, a response deviation detection and candidate anomaly space construction module 20, a cross-domain propagation topology construction and time-series tracking module 30, a fault propagation matching degree calculation module 40, a safety verification disturbance test and bilateral response acquisition module 50, and a fault source location and early warning command generation module 60.

[0072] The system includes: a multi-source test data acquisition module 10, used to acquire test control commands, engine body monitoring data, and test platform monitoring data during engine testing; a response deviation detection and candidate anomaly space construction module 20, used to perform response deviation detection on the engine body monitoring data and the test platform monitoring data according to the test control commands, and establish a candidate anomaly variable space; a cross-domain propagation topology construction and timing tracking module 30, used to construct a cross-domain variable propagation topology between the engine body and the test platform, and to perform anomaly occurrence timing tracking on the candidate anomaly variable space according to the cross-domain variable propagation topology, generating a cross-domain anomaly propagation chain; a fault propagation matching degree calculation module 40, used to calculate the fault propagation matching degree of the engine body and the test platform according to the cross-domain anomaly propagation chain; a safety verification disturbance test and bilateral response acquisition module 50, used to perform safety verification disturbance tests according to the fault propagation matching degree of the engine body and the test platform, and obtain a test bilateral response sequence; and a fault source location and early warning command generation module 60, used to locate the fault source according to the test bilateral response sequence, obtain the fault source location result, and generate a fault early warning command according to the fault source location result.

[0073] The detailed description of the specific configuration of the response deviation detection and candidate anomaly space construction module 20 is as follows: As mentioned above, response deviation detection is performed on the engine body monitoring data and the test platform monitoring data according to the test control command to establish a candidate anomaly variable space. The response deviation detection and candidate anomaly space construction module 20 may further include: a bilateral response baseline mining and feature matrix construction unit for performing bilateral response baseline mining according to the test control command to establish an engine response baseline feature matrix and a platform response baseline feature matrix; an engine body deviation detection and deviation sequence generation unit for performing response deviation detection on the engine body monitoring data according to the engine response baseline feature matrix to obtain an engine response deviation sequence; a test platform deviation detection and deviation sequence generation unit for performing response deviation detection on the test platform monitoring data according to the platform response baseline feature matrix to obtain a platform response deviation sequence; and an anomaly screening and candidate variable space generation unit for screening the engine response deviation sequence and the platform response deviation sequence according to anomaly screening conditions to generate the candidate anomaly variable space.

[0074] The bilateral response baseline mining and feature matrix construction unit, based on the test control command, performs bilateral response baseline mining to establish an engine response baseline feature matrix and a platform response baseline feature matrix. The bilateral response baseline mining and feature matrix construction unit may further include: a normal response sample set retrieval subunit for retrieving the engine normal response sample set and the platform normal response sample set according to the test control command; an engine baseline feature extraction and calibration subunit for extracting features from the engine normal response sample set to obtain an engine normal response feature set, and performing statistical aggregation and benchmark calibration on the engine normal response feature set to generate the engine response baseline feature matrix; and a platform baseline feature extraction and calibration subunit for extracting features from the platform normal response sample set to obtain a platform normal response feature set, and performing statistical aggregation and benchmark calibration on the platform normal response feature set to generate the platform response baseline feature matrix.

[0075] The detailed description of the cross-domain propagation topology construction and time-series tracking module 30 is explained below: As mentioned above, the cross-domain propagation topology is used to perform time-series tracking of anomalies in the candidate anomaly variable space to generate a cross-domain anomaly propagation chain. The cross-domain propagation topology construction and time-series tracking module 30 may further include: an anomaly variable time-series sorting and adjacency relationship acquisition unit for time-series sorting of the candidate anomaly variable space and acquiring time-series adjacency relationships of variables; an anomaly variable topology mapping and propagation edge extraction unit for mapping the candidate anomaly variable space to the cross-domain variable propagation topology based on the time-series adjacency relationships of variables and acquiring multiple anomaly variable propagation edges; a propagation edge multi-dimensional relationship identification unit for identifying the time-series continuity relationship, propagation direction relationship, and anomaly change direction relationship among the multiple anomaly variable propagation edges; and a propagation edge connection and cross-domain anomaly propagation chain generation unit for connecting the multiple anomaly variable propagation edges according to the time-series continuity relationship, the propagation direction relationship, and the anomaly change direction relationship to acquire the cross-domain anomaly propagation chain.

[0076] The detailed description of the specific configuration of the fault propagation matching degree calculation module 40 is as follows: As mentioned above, based on the cross-domain abnormal propagation chain, the engine body fault propagation matching degree and the test platform fault propagation matching degree are calculated. The fault propagation matching degree calculation module 40 may further include: a propagation chain feature extraction and target sequence generation unit for extracting features from the cross-domain abnormal propagation chain to obtain a target propagation feature sequence; a fault propagation template retrieval unit for retrieving the engine body fault propagation template and the test platform fault propagation template corresponding to the test control command; an engine template matching and multi-dimensional matching value acquisition unit for matching the target propagation feature sequence with the engine body fault propagation template to obtain a first variable sequence matching value, a first propagation direction matching value, and a first propagation delay matching value; an engine matching degree weighted fusion calculation unit for weighted fusion of the first variable sequence matching value, the first propagation direction matching value, and the first propagation delay matching value to generate the engine body fault propagation matching degree; and a test platform matching degree calculation unit for calculating the test platform fault propagation matching degree based on the target propagation feature sequence and the test platform fault propagation template.

[0077] The detailed description of the specific configuration of the safety verification disturbance test and bilateral response acquisition module 50 is explained as follows: As mentioned above, a safety verification disturbance test is performed based on the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a test bilateral response sequence. The safety verification disturbance test and bilateral response acquisition module 50 may further include: a matching discrimination difference calculation unit for calculating the difference between the engine body fault propagation matching degree and the test platform fault propagation matching degree to obtain a propagation matching discrimination degree; a dispute propagation edge identification unit for identifying dispute propagation edges based on the cross-domain abnormal propagation chain if the propagation matching discrimination degree is less than or equal to a preset discrimination threshold; a verification test scheme generation and safety optimization unit for generating an initial verification test scheme based on the dispute propagation edge, and performing safety optimization on the initial verification test scheme based on engine test safety constraints and platform test safety constraints to obtain a verification test optimization scheme; and a bilateral response parameter acquisition and sequence generation unit for acquiring the response parameters of the engine body and the test platform under the verification test optimization scheme to obtain the test bilateral response sequence.

[0078] The detailed description of the specific configuration of the fault source localization and early warning command generation module 60 is explained as follows: As mentioned above, the fault source is located based on the test bilateral response sequence to obtain the fault source localization result. The fault source localization and early warning command generation module 60 may further include: a bilateral response difference comparison and feature set generation unit for comparing response differences based on the test bilateral response sequence to generate a bilateral response difference feature set; a dual fault hypothesis expected response template fitting unit for fitting a first bilateral expected response template under the engine body fault hypothesis and a second bilateral expected response template under the test platform fault hypothesis based on the dispute propagation edge; an engine fault hypothesis registration evaluation and support calculation unit for registering and evaluating the bilateral response difference feature set with the first bilateral expected response template to obtain engine body fault support; a platform fault hypothesis registration evaluation and support calculation unit for registering and evaluating the bilateral response difference feature set with the second bilateral expected response template to obtain test platform fault support; and a support comparison and fault source localization unit for comparing the engine body fault support and the test platform fault support to generate the fault source localization result.

[0079] The cross-domain variable propagation topology between the engine body and the test platform is constructed. The cross-domain propagation topology construction and timing tracking module 30 may further include: an ontology platform relationship information collection and dataset generation unit for collecting relationship information between the engine body and the test platform and obtaining an ontology platform relationship dataset; and a cross-domain variable propagation topology construction unit for constructing the cross-domain variable propagation topology based on the ontology platform relationship dataset.

[0080] Specifically, the test platform fault propagation matching degree is calculated based on the target propagation feature sequence and the test platform fault propagation template. The test platform matching degree calculation unit may further include: a platform template matching and multi-dimensional matching value acquisition subunit for matching the target propagation feature sequence with the test platform fault propagation template to obtain a second variable sequence matching value, a second propagation direction matching value, and a second propagation delay matching value; and a platform matching degree weighted fusion calculation subunit for weighted fusion of the second variable sequence matching value, the second propagation direction matching value, and the second propagation delay matching value to generate the test platform fault propagation matching degree.

[0081] The engine test platform fault early warning system based on time-series mining provided in this embodiment of the invention can execute the engine test platform fault early warning method based on time-series mining provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0082] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A fault early warning method for engine test platforms based on time-series mining, characterized in that, The method includes: Acquire test control commands, engine body monitoring data, and test platform monitoring data during the engine testing process; Based on the test control command, response deviation detection is performed on the engine body monitoring data and the test platform monitoring data respectively to establish a candidate abnormal variable space; Construct a cross-domain variable propagation topology between the engine body and the test platform, and perform anomaly occurrence timing tracking on the candidate anomaly variable space based on the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain; Based on the cross-domain anomaly propagation chain, calculate the engine body fault propagation matching degree and the test platform fault propagation matching degree; Based on the engine body fault propagation matching degree and the test platform fault propagation matching degree, a safety verification disturbance test is performed to obtain the test bilateral response sequence. The fault source is located based on the test bilateral response sequence, the fault source location result is obtained, and a fault warning instruction is generated based on the fault source location result.

2. The engine test platform fault early warning method based on time-series mining as described in claim 1, characterized in that, Based on the test control commands, response deviation detection is performed on the engine body monitoring data and the test platform monitoring data respectively, and a candidate abnormal variable space is established, including: Based on the test control instructions, bilateral response baseline mining is performed to establish engine response baseline feature matrix and platform response baseline feature matrix; Based on the engine response baseline feature matrix, the engine body monitoring data is subjected to response deviation detection to obtain the engine response deviation sequence. Based on the platform response baseline feature matrix, the test platform monitoring data is used to detect response deviations and obtain the platform response deviation sequence. The engine response deviation sequence and the platform response deviation sequence are filtered according to the anomaly screening criteria to generate the candidate anomaly variable space.

3. The engine test platform fault early warning method based on time-series mining as described in claim 2, characterized in that, Based on the test control commands, bilateral response baseline mining is performed to establish engine response baseline feature matrices and platform response baseline feature matrices, including: According to the test control command, retrieve the engine normal response sample set and the platform normal response sample set; Feature extraction is performed on the engine normal response sample set to obtain the engine normal response feature set, and statistical aggregation and benchmark calibration are performed on the engine normal response feature set to generate the engine response baseline feature matrix. Feature extraction is performed on the platform's normal response sample set to obtain a platform normal response feature set. The platform normal response feature set is then statistically aggregated and benchmarked to generate the platform response baseline feature matrix.

4. The engine test platform fault early warning method based on time-series mining as described in claim 1, characterized in that, Based on the cross-domain variable propagation topology, the candidate abnormal variable space is subjected to anomaly occurrence time-series tracking to generate a cross-domain anomaly propagation chain, including: The candidate abnormal variable space is sorted temporally to obtain the temporal adjacency relationship of the variables; Based on the temporal adjacency relationship of the variables, the candidate abnormal variable space is mapped to the cross-domain variable propagation topology to obtain multiple abnormal variable propagation edges; Identify the temporal continuity relationship, propagation direction relationship, and abnormal change direction relationship among the multiple abnormal variable propagation edges; The cross-domain anomaly propagation chain is obtained by connecting the multiple anomaly variable propagation edges according to the temporal continuity relationship, the propagation direction relationship, and the anomaly change direction relationship.

5. The engine test platform fault early warning method based on time-series mining as described in claim 1, characterized in that, Based on the aforementioned cross-domain anomaly propagation chain, calculate the engine body fault propagation matching degree and the test platform fault propagation matching degree, including: Feature extraction is performed on the cross-domain anomaly propagation chain to obtain the target propagation feature sequence; Retrieve the engine body fault propagation template and the test platform fault propagation template corresponding to the test control command; The target propagation feature sequence is matched with the engine body fault propagation template to obtain the first variable sequence matching value, the first propagation direction matching value, and the first propagation delay matching value. The first variable sequence matching value, the first propagation direction matching value, and the first propagation delay matching value are weighted and fused to generate the engine body fault propagation matching degree. The matching degree of the test platform fault propagation is calculated based on the target propagation feature sequence and the test platform fault propagation template.

6. The engine test platform fault early warning method based on time-series mining as described in claim 1, characterized in that, Based on the engine body fault propagation matching degree and the test platform fault propagation matching degree, a safety verification disturbance test is performed to obtain the test bilateral response sequence, including: The difference between the fault propagation matching degree of the engine body and the fault propagation matching degree of the test platform is calculated to obtain the propagation matching discrimination degree. If the propagation matching discrimination is less than or equal to a preset discrimination threshold, the disputed propagation edge is identified based on the cross-domain abnormal propagation chain. Based on the dispute propagation edge, an initial verification test plan is generated, and the initial verification test plan is optimized for security based on engine test security constraints and platform test security constraints to obtain an optimized verification test plan. The response parameters of the engine body and the test platform under the verification test optimization scheme are collected to obtain the test bilateral response sequence.

7. The engine test platform fault early warning method based on time-series mining as described in claim 6, characterized in that, Based on the test bilateral response sequence, the fault source is located, and the fault source location result is obtained, including: Based on the test bilateral response sequences, a response difference comparison is performed to generate a bilateral response difference feature set; Based on the dispute propagation edge, fit a first bilateral expected response template under the engine body failure assumption and a second bilateral expected response template under the test platform failure assumption; The bilateral response difference feature set is registered and evaluated with the first bilateral expected response template to obtain the engine body fault support degree. The bilateral response difference feature set is registered and evaluated with the second bilateral expected response template to obtain the test platform fault support degree. By comparing the fault support of the engine body with the fault support of the test platform, the fault source location result is generated.

8. The engine test platform fault early warning method based on time-series mining as described in claim 1, characterized in that, Constructing the cross-domain variable propagation topology between the engine body and the test platform includes: Collect information on the relationship between the engine body and the test platform to obtain a dataset of engine body-platform relationships; Based on the ontology platform relationship dataset, construct the cross-domain variable propagation topology.

9. The engine test platform fault early warning method based on time-series mining as described in claim 5, characterized in that, Based on the target propagation feature sequence and the test platform fault propagation template, calculate the test platform fault propagation matching degree, including: The target propagation feature sequence is matched with the test platform fault propagation template to obtain the second variable sequence matching value, the second propagation direction matching value, and the second propagation delay matching value. The second variable sequence matching value, the second propagation direction matching value, and the second propagation delay matching value are weighted and fused to generate the fault propagation matching degree of the test platform.

10. A fault early warning system for an engine test platform based on time-series mining, characterized in that, The system is used to implement the engine test platform fault early warning method based on time-series mining as described in any one of claims 1-9, and the system comprises: The multi-source test data acquisition module is used to acquire test control commands, engine body monitoring data and test platform monitoring data during the test process. The response deviation detection and candidate anomaly space construction module is used to perform response deviation detection on the engine body monitoring data and the test platform monitoring data according to the test control command, and to establish a candidate anomaly variable space. The cross-domain propagation topology construction and timing tracking module is used to construct the cross-domain variable propagation topology between the engine body and the test platform, and to perform anomaly occurrence timing tracking on the candidate anomaly variable space based on the cross-domain variable propagation topology to generate a cross-domain anomaly propagation chain. The fault propagation matching degree calculation module is used to calculate the fault propagation matching degree of the engine body and the fault propagation matching degree of the test platform based on the cross-domain anomaly propagation chain. The safety verification disturbance test and bilateral response acquisition module is used to perform safety verification disturbance tests based on the engine body fault propagation matching degree and the test platform fault propagation matching degree, and to acquire the test bilateral response sequence. The fault source location and early warning command generation module is used to locate the fault source based on the test bilateral response sequence, obtain the fault source location result, and generate a fault early warning command based on the fault source location result.

Images