A method and system for monitoring emissions from an internal combustion engine in a car.

Abstract

This invention discloses a method and system for monitoring emissions from an internal combustion engine in a vehicle. The method includes: collecting a multi-dimensional fusion sensor dataset based on the operating status of the internal combustion engine; generating a dynamic emission status feature vector set representing the current combustion efficiency, wear status of key components, and emission generation paths from the multi-dimensional fusion sensor dataset; constructing and updating an emission health status map reflecting the engine's current comprehensive emission health status in real time based on the dynamic emission status feature vector set; generating an emission anomaly diagnostic report containing specific fault modes based on the emission health status map; and predicting the deterioration trend of specific emission paths of the engine based on the emission anomaly diagnostic report and historical emission health status map sequences, and generating targeted preventive engine maintenance strategies. Using this invention, real-time and accurate diagnosis and trend prediction of engine emission health status can be achieved, improving the proactiveness and maintenance efficiency of emission anomaly handling.

Description

Technical Field

[0001] This invention belongs to the field of vehicle technology, and in particular to a method and system for monitoring emissions from an internal combustion engine in a car. Background Technology

[0002] Traditional methods for monitoring emissions from automotive internal combustion engines typically rely on a single sensor (such as an oxygen sensor or exhaust gas analyzer) to determine thresholds, which fails to comprehensively reflect the complex emission conditions of the engine. Existing technologies lack coordinated analysis of combustion efficiency, component wear, and emission formation pathways, leading to delayed or misdiagnosed faults. Furthermore, discrete monitoring data cannot dynamically correlate with engine health status, nor can it predict emission deterioration trends, resulting in a lack of targeted preventative maintenance strategies. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring emissions from an internal combustion engine in a vehicle, in order to overcome the shortcomings of the prior art, and to achieve real-time and accurate diagnosis and trend prediction of the engine's emission health status, thereby improving the initiative and maintenance efficiency in handling emission anomalies.

[0004] One embodiment of this application provides a method for monitoring emissions from an internal combustion engine in a vehicle, the method comprising:

[0005] Based on the operating status of the vehicle's internal combustion engine, a multi-dimensional fusion sensor dataset is collected in real time, including exhaust gas composition spectrum data, combustion chamber pressure fluctuation waveform, cylinder temperature gradient, crankcase blow-by flow rate, and exhaust back pressure.

[0006] For the multidimensional fusion sensor dataset, spatiotemporal alignment and feature extraction are performed to generate a dynamic emission state feature vector set that characterizes the current combustion efficiency, wear status of key components, and emission generation path.

[0007] Based on the dynamic emission status feature vector set, the feature vectors are mapped to the map nodes using the pre-built emission health map model, and an emission health map reflecting the current comprehensive emission health status of the engine is constructed and updated in real time.

[0008] Based on the emission health status map, the deviation of map node status and path anomaly detection processing are performed to identify the root cause components or processes that lead to potential or actual emission exceedances, and an emission anomaly diagnosis report containing specific failure modes is generated.

[0009] Based on the emission anomaly diagnosis report and historical emission health status map sequence, a prediction algorithm based on map evolution is applied to predict the deterioration trend of specific emission paths of the engine and generate targeted preventive engine maintenance strategies.

[0010] Optionally, the step of real-time synchronously collecting a multi-dimensional fusion sensor dataset, including exhaust gas composition spectral data, combustion chamber pressure fluctuation waveform, in-cylinder temperature gradient, crankcase blow-by flow, and exhaust back pressure, based on the operating status of the vehicle's internal combustion engine, includes:

[0011] Based on the phase synchronization pulse signal output by the crankshaft position sensor, the sampling clocks of the cylinder pressure sensor, exhaust gas spectrometer, temperature array, blow-by flow meter and back pressure sensor are synchronized to obtain the original signal stream with time reference alignment;

[0012] Based on the original signal stream aligned with the time base, the combustion cycle is segmented, and the data slices for the compression stroke and power stroke are divided according to the crankshaft angle to obtain the grouped signal set of the combustion stage;

[0013] Based on the grouped signal set of the combustion stage, engine geometric space mapping is performed, and cylinder wall vibration signal is mapped to the position of piston ring-cylinder liner friction pair based on piston kinematics to obtain spatial registration vibration feature matrix;

[0014] Based on the spatially registered vibration feature matrix, multi-source noise cancellation processing is performed, exhaust gas spectrum, temperature and flow data are fused and ignition electromagnetic interference is suppressed to obtain a spatiotemporally synchronized multidimensional fused sensor dataset.

[0015] Optionally, the step of performing spatiotemporal alignment and feature extraction on the multidimensional fused sensing dataset to generate a dynamic emission state feature vector set characterizing the current combustion efficiency, wear status of key components, and emission generation paths includes:

[0016] Based on the multidimensional fusion sensor dataset, variational mode decomposition of cylinder pressure waveform is performed to extract the inflection point of pressure rise rate and the energy distribution characteristics of the indicated average effective pressure, thereby obtaining the combustion stability feature vector.

[0017] Based on exhaust gas spectral data, chemical bond absorption peaks are correlated, and the concentration ratios and transformation pathways of emissions are analyzed to obtain pollutant generation pathway feature vectors.

[0018] Based on the spatially registered vibration feature matrix, the energy attenuation gradient in the 2-5kHz frequency band is quantized using the harmonic wavelet packet entropy algorithm to obtain the piston ring wear state feature vector.

[0019] The combustion stability feature vector, pollutant generation path feature vector, and piston ring wear state feature vector are fused using tensors to construct a three-dimensional dynamic emission state tensor.

[0020] Based on the three-dimensional dynamic emission state tensor, non-negative constraint tensor decomposition is performed to extract the low-rank feature subspace representing emission degradation, thus obtaining the dynamic emission state feature vector set.

[0021] Optionally, the step of mapping the feature vectors to map nodes in the emission health map model based on the dynamic emission status feature vector set, and constructing and updating the emission health status map reflecting the current comprehensive emission health status of the engine in real time, includes:

[0022] Based on the dynamic emission status feature vector set, graph attention network matching is performed to map the nearest neighbor node in the pre-constructed emission health knowledge graph to obtain the initial node mapping relationship;

[0023] Based on the initial node mapping relationship, the node state diffusion algorithm is used to calculate the semantic similarity weight between the dynamic emission state feature vector and the node, and obtain the weighted edge connection relationship.

[0024] Based on the weighted edge connections, a time-series graph convolutional network is used to fuse historical health status sequences and update key node scores to obtain a real-time health score matrix.

[0025] Based on the real-time health score matrix, the causal chain of combustion domain-component domain-emission domain is reconstructed using the abnormal path backtracking mechanism to obtain an emission health status map that reflects the current comprehensive emission health status of the engine.

[0026] Optionally, the step of performing deviation analysis of map node states and path anomaly detection processing based on the emission health status map to identify the root cause components or processes leading to potential or actual emission exceedances, and generating an emission anomaly diagnostic report containing specific fault modes, includes:

[0027] Based on the emission health status map, the node deviation is calculated, the KL divergence between the status of key nodes and the health baseline is quantified, and the node deviation vector is obtained.

[0028] Based on the node deviation vector, the core fault propagation chain leading to excessive emissions is located using the minimum causal path search algorithm;

[0029] Based on the core fault propagation chain, thermodynamic constraints are verified to eliminate pseudo-paths that violate the temperature-pressure-flow coupling equation, thus obtaining the verified fault chain.

[0030] Based on the verified fault chain, the fault modes are decoupled to obtain a fault mode list.

[0031] Based on the failure mode list, component failure probabilities are fused, and combined with real-time data, an emission anomaly diagnostic report with confidence intervals is generated.

[0032] Optionally, based on the emission anomaly diagnostic report and historical emission health status map sequence, the step of applying a prediction algorithm based on map evolution to predict the deterioration trend of specific emission paths of the engine and generating targeted preventive engine maintenance strategies includes:

[0033] Based on the emission anomaly diagnosis report and the historical emission health status map sequence, the spatiotemporal tensor of the map evolution is constructed, and the node state time slices are stored to obtain the spatiotemporal tensor of the map evolution.

[0034] Based on the spatiotemporal tensor of the graph evolution, the spatiotemporal graph neural network is used to predict the state trajectory of key nodes within a preset time period in the future, and the predicted state trajectory is obtained.

[0035] Based on the predicted state trajectory, a three-objective Pareto optimization process is performed to find the optimal balance point between emission risk, maintenance cost and downtime, and the optimal intervention plan is obtained.

[0036] Based on the optimal intervention plan, the vehicle operation plan is integrated, and a set of preventive maintenance instructions is output from the database.

[0037] Another embodiment of this application provides an emission monitoring system for an automotive internal combustion engine, the system comprising:

[0038] The acquisition module is used to collect multi-dimensional fusion sensor datasets in real time, including exhaust gas composition spectrum data, combustion chamber pressure fluctuation waveform, cylinder temperature gradient, crankcase blow-by flow rate, and exhaust back pressure, based on the operating status of the automotive internal combustion engine.

[0039] The extraction module is used to perform spatiotemporal alignment and feature extraction on the multidimensional fused sensing dataset to generate a dynamic emission state feature vector set that characterizes the current combustion efficiency, wear status of key components, and emission generation path.

[0040] The construction module is used to map the feature vectors to the map nodes based on the dynamic emission status feature vector set and a pre-built emission health map model, and to build and update the emission health status map that reflects the current comprehensive emission health status of the engine in real time.

[0041] The detection module is used to perform deviation analysis of the status of the nodes in the emission health status map and path anomaly detection processing based on the emission health status map, identify the root cause components or processes that lead to potential or actual emission exceedances, and generate an emission anomaly diagnostic report containing specific fault modes.

[0042] The prediction module is used to predict the deterioration trend of specific emission paths of the engine based on the emission anomaly diagnosis report and the historical emission health status map sequence, and to generate targeted preventive engine maintenance strategies.

[0043] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0044] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0045] Compared with existing technologies, the present invention provides an emission monitoring method for automotive internal combustion engines. This method involves real-time synchronous acquisition of a multi-dimensional fusion sensor dataset based on the engine's operating status; generating a dynamic emission status feature vector set representing current combustion efficiency, wear status of key components, and emission generation paths from the multi-dimensional fusion sensor dataset; constructing and updating an emission health status map reflecting the engine's current comprehensive emission health status in real time based on the dynamic emission status feature vector set; generating an emission anomaly diagnostic report containing specific fault modes based on the emission health status map; and predicting the deterioration trend of specific emission paths based on the emission anomaly diagnostic report and historical emission health status map sequences, and generating targeted preventative engine maintenance strategies. This enables real-time and accurate diagnosis and trend prediction of engine emission health status, improving the proactiveness and maintenance efficiency of emission anomaly handling. Attached Figure Description

[0046] Figure 1 A hardware structure block diagram of a computer terminal for an emission monitoring method for an internal combustion engine of an automobile, provided in an embodiment of the present invention;

[0047] Figure 2 A schematic flowchart illustrating an emission monitoring method for an internal combustion engine of an automobile provided by an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of the structure of an emission monitoring system for an internal combustion engine of an automobile, provided as an embodiment of the present invention. Detailed Implementation

[0049] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0050] This invention first provides a method for monitoring emissions from an internal combustion engine in a vehicle. This method can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0051] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for an emission monitoring method for an internal combustion engine in an automobile, provided as an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0052] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any emission monitoring method for an internal combustion engine in a vehicle.

[0053] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0054] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When executed by a processor, the computer program can enable the processor to perform any emission monitoring method for an internal combustion engine in a car.

[0055] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0056] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0057] See Figure 2 The present invention provides an emission monitoring method for an internal combustion engine of an automobile, which may include the following steps:

[0058] S201 collects multi-dimensional fusion sensor datasets in real time, including exhaust gas composition spectrum data, combustion chamber pressure fluctuation waveform, cylinder temperature gradient, crankcase blow-by flow rate, and exhaust back pressure, based on the operating status of the automotive internal combustion engine.

[0059] Specifically, the sampling clocks of the cylinder pressure sensor, exhaust gas spectrometer, temperature array, blow-by flow meter, and back pressure sensor can be synchronized with the phase synchronization pulse signal output by the crankshaft position sensor to obtain the original signal stream with time reference alignment.

[0060] The core synchronization mechanism of the Engine Control Unit (ECU) relies on the Crankshaft Position Sensor (CPS). This sensor, mounted at the end of the crankshaft or near the flywheel, generates a phase synchronization pulse signal (PSPS) every time the crankshaft rotates a specific angle (e.g., 6 degrees or 1 degree) using magnetoresistive or Hall effect principles. The rising or falling edge of this pulse signal corresponds to the precise physical position of the crankshaft angle (e.g., 30 degrees before top dead center). The ECU has a dedicated Clock Synchronization Module (CSM) that uses the PSPS as a Global Time Reference (GTR) to send a unified Sampling Trigger Command (STC) to all sensors. For example, the Cylinder Pressure Sensor (CyPS), mounted in the combustion chamber spark plug hole or a dedicated pressure test port, originally operates its sampling clock (SCLK) independently at a 1 MHz frequency. When the CSM detects the rising edge of the PSPS, it immediately sends an STC signal to the CyPS, forcing its next sampling point to align with the crankshaft angle. The sampling period of the Exhaust Gas Spectrometer (EGS) is typically in the millisecond range. The CSM calculates the EGS's equivalent sampling time window (STW) based on the PSPS interval (e.g., one pulse every 6 degrees) to ensure that the start time of each spectral scan is strictly synchronized with the crankshaft angle.

[0061] The Temperature Array (TA) consists of multiple miniature thermocouples (TCs) or resistance temperature detectors (RTDs), distributed at key locations in the cylinder head, cylinder wall, or exhaust manifold. Each probe has an independent sampling clock, and the CSM (Cyclic Streaming Meter) achieves synchronization via a Distributed Timestamp Protocol (DTP): when the PSPS pulse arrives, the CSM broadcasts a timestamped synchronization frame (SF) to all TA nodes. Each node dynamically adjusts its sampling instant (SI) based on the clock skew (CSK) deviation between its local clock and the TS. Blow-by flow meters (BFMs) typically employ hot-film or ultrasonic principles, and their flow sampling period is related to the crankshaft speed. The CSM dynamically calculates the BFM sampling interval (SIV) based on the PSPS frequency (e.g., a pulse interval of 33.3 microseconds at 3000 rpm), ensuring that each flow data point corresponds to a fixed crankshaft angle range (e.g., sampling once every 10 degrees). The back-pressure sensor (BPS) is installed at the end of the exhaust manifold, and its signal is susceptible to exhaust pulsation interference. The CSM utilizes the PSPS to trigger the BPS's sample-and-hold circuit (SHC), collecting effective pressure values ​​only during stable phases when the exhaust valve is closed (e.g., crankshaft angles of 240-480 degrees), thus avoiding pulsation noise.

[0062] Ultimately, all sensor data is transmitted to the ECU via a Controller Area Network (CAN) bus or a dedicated high-speed data link (such as Ethernet). The ECU's Signal Alignment Buffer (SAB) allocates an independent buffer for each sensor, and the data is sorted by PSPS timestamp index (TIDX). For example, after the Nth PSPS pulse is triggered, CyPS uploads the cylinder pressure value (in bar) within 0.1 milliseconds, EGS uploads the spectral data (absorbance matrix) for that window within 5 milliseconds, TA uploads the temperature at each point (in degrees Celsius) within 2 milliseconds, BFM uploads the blow-by flow rate value (in liters per minute), and BPS uploads the back pressure value (in kilopascals). The SAB packages all data triggered by the same pulse into a Time-Base Aligned Raw Signal Stream (TBARSS) indexed by TIDX. Each data packet in this stream contains a Global Time Tag (GTT) accurate to the microsecond level, ensuring the comparability of all physical quantities in subsequent analyses.

[0063] Based on the original signal stream aligned with the time base, the combustion cycle is segmented, and the data slices for the compression stroke and power stroke are divided according to the crankshaft angle to obtain the grouped signal set of the combustion stage;

[0064] The ECU's Cycle Segmentation Engine (CSE) parses the TBARSS data stream, first identifying the boundaries of the engine cycle (EC). Each complete cycle corresponds to 720 degrees of crankshaft rotation (for a four-stroke engine), marked by the Top Dead Center Signal (TDCS). The TDCS is marked by a specific pulse in the PSPS sequence (e.g., the 0-degree crankshaft angle pulse). Based on the TDCS, the CSE divides the TBARSS into consecutive independent Cycle Data Blocks (CDBs). Each CDB contains all sensor data packets within that cycle, ordered by TIDX. Subsequently, the CSE further subdivides each CDB into four strokes using the Crankshaft Angle Resolver (CAR): intake stroke (0-180 degrees), compression stroke (180-360 degrees), power stroke (360-540 degrees), and exhaust stroke (540-720 degrees). Compression and power strokes are the core of combustion analysis.

[0065] For the compression stroke (CS), CSE extracts a subset of data from the crankshaft angle of 180-360 degrees. In this range, cylinder pressure data (CyPS) shows a monotonically increasing trend, and the temperature array (TA) shows an increasing in-cylinder gas temperature gradient. EGS typically does not collect data during this stage (because the exhaust valves are closed). CSE resamples the CS range data (Resampling, RS) according to a preset angle resolution (e.g., every 1 degree), generating uniformly spaced compression stroke data slices (CSDS). This slice contains key parameters such as the cylinder pressure curve, cylinder wall temperature distribution, and crankcase blow-through flow (BFM). Data processing for the power stroke (PS) is more complex: CSE extracts data from 360-540 degrees and identifies combustion feature points (CFPs), such as the peak cylinder pressure position (usually 10-20 degrees after top dead center). Based on CFP, CSE subdivides PS into pre-combustion period (360 degrees to spark ignition), main combustion period (ignition to peak cylinder pressure), and aftercombustion period (peak to 540 degrees), generating high-resolution power stroke data slices (PSDS), including details of cylinder pressure fluctuations, transient emission spectra of EGS, and evolution of in-cylinder temperature field.

[0066] Finally, CSE packages the CSDS and PSDS for each engine cycle into a Combustion Phase Grouped Signal Set (CPGSS). This set is a structured data container, for example:

[0067] Compression stroke group: includes cylinder pressure array (CyPS_Array) for every 1 degree within 180-360 degrees, cylinder head temperature distribution map (TA_Map), and blow-by flow sequence (BFM_Seq);

[0068] Power stroke group: includes cylinder pressure micro-fluctuation waveforms (CyPS_Wave) every 0.5 degrees within 360-540 degrees, nitrogen oxide (NOx) spectral absorption peaks of EGS (EGS_NOxPeak), and coordinates of in-cylinder high-temperature hot spots (TA_HotSpot).

[0069] All data are accompanied by crankshaft angle labels (CAL) for easy subsequent correlation analysis. CPGSS data is stored by cyclic numbering, forming a database with a two-dimensional time-angle index.

[0070] Based on the grouped signal set of the combustion stage, engine geometric space mapping is performed, and cylinder wall vibration signal is mapped to the position of piston ring-cylinder liner friction pair based on piston kinematics to obtain spatial registration vibration feature matrix;

[0071] The cylinder wall vibration signal (CWVS) is collected by a vibration accelerometer (VA) mounted on the outside of the cylinder block; its raw data only reflects the overall vibration energy. To achieve wear localization, engine geometric space mapping (EGSM) is required. The EGSM module first loads the engine's 3D geometric model (GM), including parameters such as the cylinder bore diameter (CBD), piston pin height (PPH), and connecting rod length (CRL). Based on the piston kinematics equation (PKE), the functional relationship between the piston instantaneous position (PIP) and the crankshaft angle (CAL) is calculated. For example, when CAL = 360 degrees (top dead center), PIP is at its highest point; when CAL = 540 degrees (bottom dead center), PIP is at its lowest point.

[0072] The EGSM module parses the CWVS data (typically sampled at 50 kHz) from CPGSS and extracts vibration wave segments (VWS) by CAL slices. Based on the PIP calculation results, the Piston Ring-Liner Contact Zone (PRLCZ) corresponding to each CAL is determined. For example, when the piston is at top dead center, the coordinate of the top of the cylinder liner corresponding to the first piston ring is Z=Max; when the piston descends to a crankshaft angle of 400 degrees, the position of this ring is Z=Max -30 mm (the specific value is calculated by PKE). Subsequently, EGSM models the spatial relationship between the physical installation position of the VA (e.g., at cylinder height Z=80 mm) and the PRLCZ as a transfer function matrix (TFM). The TFM is solved using the inverse convolution algorithm (ICA), back-projecting (BP) the energy components in the original vibration signal to the actual occurrence position of the PRLCZ.

[0073] The spatially registered vibration feature matrix (SRVFM) generated after projection is a two-dimensional structure:

[0074] Horizontal dimension: Cylinder liner axial height coordinate (Z-Axis Coordinate, ZAC), resolution 1 mm, range covering piston stroke (e.g., 0-100 mm).

[0075] Column dimension: Vibration Feature Parameters (VFP), including time domain parameters (root mean square value RMS, crest factor) and frequency domain parameters (energy in the 2-5 kHz band, harmonic distortion rate THD).

[0076] For example, the abnormally high RMS value of SRVFM at Z=45 mm indicates piston ring scraping wear at that height; the 5 kHz energy decay at Z=10 mm suggests microcracks at the cylinder liner tip. This matrix transforms the vibration signal from "overall noise" into "spatially located wear fingerprints".

[0077] Based on the spatially registered vibration feature matrix, multi-source noise cancellation processing is performed, exhaust gas spectrum, temperature and flow data are fused and ignition electromagnetic interference is suppressed to obtain a spatiotemporally synchronized multidimensional fused sensor dataset.

[0078] The core of multi-source noise cancellation (MNC) is to separate the effective wear component from the interference noise in the cylinder wall vibration signal. The main noise sources include:

[0079] Ignition Electromagnetic Interference (IEMI): High-frequency radiation (frequency band >100 MHz) generated by spark plug discharge is coupled to the vibration sensor circuit.

[0080] Valve Impact Noise (VIN): The mechanical impact of the intake and exhaust valves when they come to rest (frequency band 1-3 kHz).

[0081] Crankshaft bearing vibration (CBV): low-frequency broadband vibration (frequency band <800 Hz).

[0082] The MNC module employs a Reference Sensor Cooperative Filtering (RSCF) strategy: an additional Ignition Current Probe (ICP) is deployed to acquire IEMI waveforms, an Acoustic Emission Sensor (AES) is installed in the cylinder head to capture VIN characteristics, and a Low-Frequency Accelerometer (LFA) is installed in the crankcase to monitor CBV.

[0083] For vibration data in SRVFM, MNC performs three steps of processing:

[0084] IEMI suppression: Cross-correlation analysis (CCA) is performed on the interference waveform (Ignition Noise Template, INT) acquired by ICP and the vibration signal. After calculating the time delay compensation, it is adaptively subtracted from the original signal (Adaptive Subtraction, AS).

[0085] VIN / CBV separation: A blind source separation (BSS) algorithm is used, with AES and LFA signals as reference inputs, to perform independent component analysis (ICA) on the vibration signal to extract the pure piston ring friction component.

[0086] Frequency band enhancement: Enhance the 2-5 kHz frequency band (piston ring wear characteristic band) through wavelet threshold denoising (WTD) and suppress energy in other frequency bands.

[0087] The purified vibration characteristics were spatiotemporally fused with other data from CPGSS:

[0088] Exhaust gas spectral data (EGS): By aligning the NOx absorption peak area (unit: mV·s) with the cylinder pressure curve of the power stroke, a correlation between combustion temperature and emission formation is established.

[0089] Temperature Array Data (TA): Extract the axial temperature gradient of the cylinder liner (unit: degrees Celsius / mm), and superimpose it with the RMS value of the same location vibration from SRVFM to identify wear hotspots caused by overheating.

[0090] Blow-through flow rate (BFM): Peak vibrational energy corresponding to the flow rate surge (unit: L / min) and piston ring leakage in the compression stroke slice.

[0091] The final result is a spatiotemporally synchronized multidimensional fusion sensor dataset. This dataset has a five-dimensional structure: time dimension (crankshaft angle), spatial dimension (cylinder liner height), sensing dimension (pressure / vibration / temperature / spectrum / flow), cycle dimension (engine cycle number), and feature dimension (physical quantity + statistical indicators), providing complete input for subsequent emissions health analysis.

[0092] This step involves the collaborative acquisition of key physicochemical parameters during engine operation using multiple sensors, including exhaust gas composition characteristics obtained by a spectrometer, combustion dynamics recorded by a high-frequency cylinder pressure sensor, in-cylinder thermal field distribution measured by a distributed temperature probe, blow-by volume monitored by a flow meter, and exhaust resistance characteristics captured by a back pressure sensor. All data is synchronized at the microsecond level via crankshaft phase signals, ensuring spatiotemporal consistency and overcoming the limitations of traditional single-point monitoring. This constructs a multi-dimensional data foundation reflecting the engine's true operating state, providing comprehensive and synchronized raw data support for subsequent accurate diagnosis and effectively resolving analytical errors caused by asynchronous sampling.

[0093] S202, For the multi-dimensional fusion sensing dataset, perform spatiotemporal alignment and feature extraction to generate a dynamic emission state feature vector set that characterizes the current combustion efficiency, wear status of key components and emission generation path;

[0094] Specifically, based on the multidimensional fusion sensor dataset, variational mode decomposition of cylinder pressure waveform can be performed to extract the inflection point of pressure rise rate and the energy distribution characteristics of the indicated average effective pressure, thereby obtaining the combustion stability feature vector.

[0095] The system receives a multidimensional fusion sensor dataset (MFSD) from the pre-processing stage. This dataset has been spatiotemporally aligned to ensure the synchronization of signals such as combustion chamber pressure fluctuation waveforms (cylinder pressure waveforms) and exhaust gas spectra. The cylinder pressure waveform is the core input, acquired by a high-frequency cylinder pressure sensor (typically with a sampling rate of 100 kHz, i.e., 100,000 samples per second) within each combustion cycle, and is represented as a pressure curve that varies with the crankshaft angle. To analyze combustion stability, the original cylinder pressure waveform is first processed using the Variational Mode Decomposition (VMD) algorithm. VMD decomposes the non-stationary cylinder pressure signal into several intrinsic mode functions (IMFs) through the principle of adaptive frequency segmentation. In practice, the modal number K is set to 5 (empirical value), the penalty factor α is set to 2000 (control bandwidth), and the cylinder pressure signal of a single cycle is iteratively optimized and solved to finally separate the sub-signal components representing different physical processes: such as IMF1 corresponding to high-frequency combustion oscillation, IMF2 reflecting the main combustion pressure wave, and IMF3 containing information on pressure rise rate, etc.

[0096] In the decomposed modes, IMF2 and IMF3 are analyzed in detail. The extraction process of the Pressure Rise Rate Inflection Point (PRR_IP) is as follows: the first derivative of the IMF3 component is calculated to obtain the pressure rise rate curve, and the sliding window standard deviation algorithm is used to detect its abrupt change position. When the slope change of 5 consecutive sampling points (corresponding to 0.1 degrees of crankshaft rotation angle) exceeds the threshold Δslope = 0.5 bar / degree, it is determined to be an inflection point. This inflection point marks the timing of the transition from premixed combustion to diffusion combustion. If the angle of its occurrence deviates from the standard value (e.g., 8 degrees after top dead center) by more than ±2 degrees, it is considered combustion instability. Simultaneously, the Indicated Mean Effective Pressure Energy Distribution Feature (IMEP_EDF) is extracted from the IMF2 component: the integral area ratio of IMF2 during the compression stroke (crankshaft angle -180° to 0°) and the power stroke (0° to 180°) is calculated, denoted as the work ratio (WR); then, the energy ratio (ER) of the IMF2 signal in the 3-10 kHz frequency band is calculated. WR reflects the effective work efficiency, and ER characterizes the pressure oscillation intensity; these two constitute the core dimensions of IMEP_EDF.

[0097] The final output Combustion Stability Feature Vector (CSFV) is a five-dimensional array: [PRR_IP angle deviation, WR value, ER value, maximum pressure rise rate, pressure cycle variation coefficient]. The pressure cycle variation coefficient is calculated using the standard deviation of cylinder pressure peak values ​​over 50 consecutive cycles (unit: percentage %). For example, under a certain operating condition, the vector might be: [-1.5, 0.82, 0.15, 4.3, 3.8%], representing an inflection point delay of 1.5 degrees, a power efficiency of 82%, a high-frequency oscillation energy ratio of 15%, a maximum pressure rise rate of 4.3 bar / degree, and a cycle variation rate of 3.8%. This vector is transmitted in real-time to the subsequent tensor fusion module, and any abnormalities in its values ​​can be directly correlated with combustion problems such as ignition delay and air-fuel ratio imbalance.

[0098] Based on exhaust gas spectral data, chemical bond absorption peaks are correlated, and the concentration ratios and transformation pathways of emissions are analyzed to obtain pollutant generation pathway feature vectors.

[0099] Exhaust Gas Spectrum Data (EGSD) was acquired from a UV-IR broadband spectrometer (wavelength coverage from 200 nm to 5000 nm) at a rate of 100 frames per second. The raw spectra were first baseline corrected: scattering noise was eliminated using Adaptive Iteratively Reweighted Penalized Least Squares (AIRPLS), and the optical path effect was normalized using Standard Normal Variation (SNV). The preprocessed spectra were then input into a chemical bond absorption peak correlation module, which incorporates a database of characteristic pollutant peaks. For example, nitric oxide (NO) has a strong absorption band at 5.3 μm (mid-infrared), the CH bond stretching vibration peak of hydrocarbons (HC) is located at 3.4 μm, and nitrogen oxides (NO2) have a characteristic absorption at 400 nm (UV).

[0100] Emission concentration analysis was performed using a partial least squares regression (PLSR) model. A mapping relationship between the spectral absorbance matrix and the measured concentrations by gas chromatograph (GC) was established beforehand through bench testing. During real-time operation, the absorbance sequence of the target spectrum in the characteristic band (e.g., NO: 5.2-5.4 μm) was input into the PLSR model, which outputs the concentration values ​​(in parts per million, ppm) of NO, HC, CO, CO2, and NO2. The conversion pathway analysis relies on the extraction of reaction kinetic features: the NOx Formation Index (NFI) is calculated as (NO concentration + NO2 concentration) / (peak in-cylinder temperature × oxygen concentration). An index exceeding the threshold of 0.25 indicates that thermal NOx is dominant. At the same time, the peak asymmetry index (AI) of the HC spectrum in the 2900-3000 wavenumber cm⁻¹ is analyzed. If AI > 1.2, it is determined that there are unburned fuel droplets, which are HC emissions caused by the wet wall effect.

[0101] The generated Pollutant Generation Path Feature Vector (PGPFV) contains a six-tuple: [NO concentration, HC concentration, NOx generation index, HC asymmetry, soot transmittance, ammonia-to-NOx ratio]. Soot transmittance is calculated from the transmittance at a wavelength of 660 nm (unit: percentage %), reflecting the particulate matter generation level. The ammonia-to-NOx ratio (ANR) = (theoretical ammonia yield corresponding to urea injection rate) / (NO concentration × exhaust flow rate), used to monitor the efficiency of the selective catalytic reduction system. For example, the vector [120, 35, 0.31, 1.5, 88%, 0.9] indicates: NO 120 ppm, HC 35 ppm, predominantly thermal NOx, strong wet-wall effect, moderate soot concentration, and insufficient urea injection rate. This vector provides a chain of evidence for subsequent diagnostics based on the chemical reaction dimension.

[0102] Based on the spatially registered vibration feature matrix, the energy attenuation gradient in the 2-5kHz frequency band is quantized using the harmonic wavelet packet entropy algorithm to obtain the piston ring wear state feature vector.

[0103] The Spatially Registered Vibration Feature Matrix (SRVFM) originates from the pre-processing stage. Its rows correspond to the engine cylinder number (e.g., 4 rows for a 4-cylinder engine), and the columns represent the crankshaft angle position (0.5-degree resolution). Matrix elements are the vibration acceleration values ​​(unit: gravitational acceleration g) at specific cylinder-to-cylinder angle points. This matrix focuses on the piston ring-liner friction pair (PRLFP) region, mapping cylinder wall vibration sensor signals to the piston ring top dead center / bottom dead center positions using a piston kinematics model. For example, the vibration acceleration at the top dead center position of cylinder 2 (crankshaft angle 0°) is extracted as the key observation point.

[0104] The Harmonic Wavelet Packet Entropy (HWPE) algorithm is executed in three steps:

[0105] Step 1 - Harmonic Wavelet Packet Decomposition: For the vibration signal at each key location (such as the top dead center), perform 7-level wavelet packet decomposition using harmonic wavelet basis functions (center frequency adjustable). Select the 2-5 kHz frequency band (corresponding to the 4th-5th sub-bands), as this frequency band is sensitive to piston ring wear.

[0106] Step 2 - Energy Attenuation Gradient Calculation: Within the target frequency band, calculate the energy value (EV) according to the crankshaft angle interval. For example, divide the compression stroke (-30° to 0°) into 10 equal segments, and calculate the sum of squared wavelet packet coefficients for each segment as the EV. Energy Attenuation Gradient (EAG) = (EV of segment 1 - EV of segment 10) / (angle span). As wear intensifies, lubrication deteriorates, causing the high-frequency vibration energy to attenuate more slowly in the later stages of the compression stroke, thus reducing the absolute value of EAG.

[0107] Step 3 - Entropy Analysis: Calculate the permutation entropy (PE) of the wavelet packet coefficients in the target frequency band to measure signal complexity. Vibration randomness increases after piston ring wear, leading to a rise in the PE value.

[0108] The output Piston Ring Wear State Feature Vector (PRWSFV) contains four parameters: [energy attenuation gradient in the 2-5kHz frequency band at top dead center, entropy of the same frequency band at bottom dead center, total energy value of the entire stroke frequency band, and wear consistency index]. The Wear Consistency Index (WCI) = (standard deviation of EAG for each cylinder) / (mean), used to detect uneven wear. For example, the vector [-0.25, 0.78, 15.3, 0.18] indicates: gentle energy attenuation at top dead center (gradient -0.25 g² / degree), high vibration complexity at bottom dead center (PE=0.78), total high-frequency energy value of 15.3 g², and wear difference of 18% among cylinders. This vector can directly quantify the degree of piston ring sealing degradation.

[0109] The combustion stability feature vector, pollutant generation path feature vector, and piston ring wear state feature vector are fused using tensors to construct a three-dimensional dynamic emission state tensor.

[0110] The three input feature vectors are strictly spatiotemporally aligned: all are generated based on the same combustion cycle. The combustion stability feature vector (CSFV) has a dimension of 5 (pressure-related parameters), the pollutant formation path feature vector (PGPFV) has a dimension of 6 (emission chemical parameters), and the piston ring wear state feature vector (PRWSFV) has a dimension of 4 (mechanical wear parameters). The system constructs a feature matrix (FM) for each cycle: rows correspond to feature categories (3 rows in total), and columns expand to specific parameters. For example, the first row contains 5 parameters for CSFV, the second row contains 6 parameters for PGPFV (padded with zeros if necessary), and the third row contains 4 parameters for PRWSFV.

[0111] Tensor fusion is achieved through the expansion of higher-order tensors. The three dimensions of the 3D Dynamic Emission State Tensor (3D-DEST) are defined as follows:

[0112] Dimension 1 (Combustion Dimension): Contains all 5 elements of CSFV, characterizing the state of thermodynamic processes;

[0113] Dimension 2 (Emissions Dimension): Contains all 6 elements of PGPFV, characterizing the chemical reaction state;

[0114] Dimension 3 (Component Dimension): Contains all four elements of PRWSFV, characterizing the mechanical wear state.

[0115] Tensor element values ​​are processed through feature standardization: the mean and standard deviation of the most recent 1000 cycles are calculated independently for each feature parameter, and Z-score standardization is used: (current value - mean) / standard deviation. For example, if the original value of the cylinder pressure cycle variation coefficient is 3.8%, and the historical mean is 4.0% and the standard deviation is 0.5%, then the standardized value is (3.8 - 4.0) / 0.5 = -0.4.

[0116] The final constructed three-dimensional dynamic emission state tensor (3D-DEST) has a size of 5×6×4 (a total of 120 elements). Its physical significance lies in establishing cross-domain correlations: for example, the tensor element (3,2,1) represents the coupling state between "maximum pressure rise rate (the third parameter of the combustion dimension)" and "NO concentration (the second parameter of the emission dimension)" under the condition of "top dead center energy decay gradient (the first parameter of the component dimension)". This tensor is updated in real time (once per cycle) and input into the subsequent feature decomposition module, providing a structured data foundation for global emission health analysis.

[0117] Based on the three-dimensional dynamic emission state tensor, non-negative constraint tensor decomposition is performed to extract the low-rank feature subspace representing emission degradation, thus obtaining the dynamic emission state feature vector set.

[0118] Non-negative Tensor Factorization (NTF) employs the PARAFAC (Parallel Factor Analysis) model. The three-dimensional tensor 3D-DEST is decomposed into three factor matrices (combustion factor matrix, emission factor matrix, and component factor matrix) and a core tensor. The decomposition rank R is set to 3 (an empirical value), meaning three latent feature patterns are extracted. The decomposition process forces all factor elements to be ≥0, conforming to the physical meaning of engineering parameters (e.g., concentration and wear have no negative values). The optimization algorithm uses Alternating Least Squares (ALS), with an upper limit of 100 iterations and a convergence threshold of 1e-6.

[0119] Analysis focusing on the core tensor extracted from the low-rank feature subspace. The core tensor has a size of 3×3×3, and its element values ​​represent the coupling strength between factors. The Emission Degradation Indicating Subspace (EDIS) is defined as the subspace consisting of the second column of the combustion factor matrix (associated with unstable combustion), the first column of the emission factor matrix (associated with NOx generation), and the third column of the component factor matrix (associated with wear consistency). The energy ratio (EER) of this subspace is calculated as (sum of squares of the core tensor elements in the subspace) / (sum of squares of the entire core tensor elements). A significant degradation mode is considered to exist when EER > 0.35.

[0120] The output Dynamic Emission State Feature VectorSet (DESFVS) contains three vectors:

[0121] Vector 1 (combustion factor vector): The 5 elements in the second column of the combustion factor matrix represent the projected weights of the unstable combustion mode on each pressure parameter;

[0122] Vector 2 (emission factor vector): The 6 elements in the first column of the emission factor matrix represent the distribution of the NOx-dominant mode in terms of chemical parameters;

[0123] Vector 3 (Component Factor Vector): The four elements in the third column of the component factor matrix represent the manifestation of wear inconsistency in mechanical parameters.

[0124] For example, the emission factor vector might be [0.9, 0.1, 0.7, 0.3, 0.2, 0.05], indicating that NO concentration (weight 0.9) and NOx formation index (weight 0.7) are the main causes of this degradation pattern. This vector set serves as input to the emissions health profile model, enabling the transformation from raw data to interpretable features.

[0125] Advanced signal processing methods such as variational mode decomposition and harmonic wavelet packet entropy are employed to extract characteristic indicators with clear physical meaning from the raw data. These include combustion stability reflected by the inflection point of pressure rise rate, piston ring wear characterized by vibration energy decay in a specific frequency band, and the trend of spectral absorption peak changes related to NOx generation pathways. The raw data is transformed into a quantitative feature set that can be directly used for condition assessment. Through feature engineering, the coupling relationship between combustion efficiency, mechanical wear, and emission generation hidden behind the data is revealed, laying the foundation for establishing an accurate emission health model.

[0126] S203, Based on the dynamic emission status feature vector set, the feature vectors are mapped to the map nodes using the pre-constructed emission health map model, and an emission health map reflecting the current comprehensive emission health status of the engine is constructed and updated in real time.

[0127] Specifically, graph attention network matching can be performed based on the dynamic emission status feature vector set to map the nearest neighbor node in the pre-constructed emission health knowledge graph and obtain the initial node mapping relationship;

[0128] Structure and initialization of pre-constructed emissions health knowledge graph

[0129] The Emission Health Knowledge Graph (EHKG) is a pre-generated structured knowledge base where nodes represent entities related to engine emissions (such as the combustion chamber, piston rings, three-way catalytic converter, and nitrogen oxide generation pathways), and edges represent causal relationships between entities (such as "piston ring wear → cylinder pressure leakage → increased unburned hydrocarbons"). Each node stores a multi-dimensional attribute vector, including historical health status baseline values ​​(such as normal cylinder pressure fluctuation range of 0.5-1.2 MPa), fault mode association weights (such as the correlation coefficient between piston ring wear and hydrocarbon emissions of 0.85), and physical constraints (such as the coefficients of the temperature-pressure coupling equation). The graph is jointly constructed using an expert knowledge base and a historical fault database. For example, the OBD fault code P0420 (low catalytic converter efficiency) is mapped to the "catalytic converter aging" node, and bidirectional edge connections are established with the "oxygen sensor failure" and "air-fuel ratio mismatch" nodes. During graph initialization, the health baseline vector of each node is trained using tens of thousands of sets of normal operating condition data, serving as a reference for subsequent matching.

[0130] The core computational process of graph attention network matching

[0131] A dynamic emission state feature vector set (e.g., a 50-dimensional vector containing combustion stability features, piston ring wear features, and NOx path features) is input into a Graph Attention Network (GAT). GAT first calculates the similarity score (SS) between the input feature vector and each node attribute in the EHKG. The equivalent operation is as follows: For each node, a linear transformation is performed on the input features and node features using a learnable weight matrix. Then, the attention mechanism is used to calculate the cosine similarity between the two in the latent space, and the influence weights of adjacent nodes are added. For example, for the "cylinder pressure leakage" node, GAT simultaneously calculates its association strength with the adjacent "piston ring wear" and "valve seal failure" nodes, ultimately outputting a matching confidence (MC, range 0-1). The system sets an MC threshold of 0.7; a match is considered valid only when MC ≥ 0.7. After matching, the node with the highest confidence is selected as the Nearest Neighbor Node (NNN). For example, if abnormal cylinder pressure fluctuations are detected (characteristic value deviates from the baseline by 30%) and accompanied by an increase in crankcase blow-by flow, GAT may map the input vector to the "increased piston ring-cylinder liner clearance" node, MC=0.92.

[0132] Generation and verification of initial node mapping relationships

[0133] The matching results generate an Initial Node Mapping Table (INMT), recording the correspondence between dynamic feature vectors and EHKG nodes (e.g., feature vector ID-7 maps to the node "Catalyst oxygen storage capacity decline"). To ensure mapping reliability, the system initiates a cross-modal verification mechanism: for example, when GAT maps a feature vector to the "Injector clogging" node, the verification module compares the carbon soot absorption peak intensity (2.5-micron absorbance > 0.8) in the real-time exhaust gas spectrum with the clogging feature threshold stored in the node; if they do not match, a secondary matching is triggered. The final output is a mapping relationship with confidence labels for subsequent processing.

[0134] Based on the initial node mapping relationship, the node state diffusion algorithm is used to calculate the semantic similarity weight between the dynamic emission state feature vector and the node, and obtain the weighted edge connection relationship.

[0135] Physical basis and algorithm initialization of node state propagation

[0136] The Node State Diffusion Algorithm (NSDA) is designed based on the fault propagation characteristics within an engine system: for example, piston ring wear not only directly affects blow-by but also indirectly increases nitrogen oxide generation by reducing the compression ratio. The algorithm starts with the initially mapped node and performs multi-hop diffusion (MHD) along the edges of the EHKG. The diffusion depth is controlled by a preset propagation order (PO, usually set to 3), meaning it calculates up to the third level of neighbors of the target node. The diffusion weight initialization follows the thermodynamic decay principle: edges directly connected to the fault source are assigned a base weight of 0.9, and the weight decays by 0.3 for each additional hop (e.g., second-order edge weight = 0.9 × 0.3 = 0.27). The algorithm also considers edge type weights (e.g., mechanical wear edges have a weight coefficient of 1.0, chemical reaction edges 0.8).

[0137] Dynamic calculation of semantic similarity weights

[0138] For each diffusion path, the semantic similarity weight (SSW) between the dynamic feature vector and the path's terminal node is calculated. This weight is achieved through bidirectional feature alignment.

[0139] Forward matching: Extract key parameters from the feature vector (such as cylinder pressure rise rate of 0.15 MPa / degree) and calculate the overlap ratio (OR) with the fault parameter range stored in the node (such as normal range of 0.18-0.22).

[0140] Reverse derivation: Based on the physical constraint equations defined by the node (such as the modified ideal gas law), reverse derivation is made to determine whether the current feature vector satisfies the node causal relationship (for example, if the exhaust back pressure exceeds the limit by 1.8 bar, the SSW of the "catalytic converter blockage" node increases by 0.2).

[0141] Finally, the SSW is equivalently combined with the OR and reverse verification scores according to the formula. For example, the SSW of a certain feature vector and the "oxygen sensor aging" node is 0.6 (OR contribution) + 0.3 (reverse verification) = 0.9.

[0142] Generation of weighted edge connections

[0143] The algorithm outputs a weighted edge connection matrix (WECM). Rows represent the initial mapped nodes, columns represent the spread-cover nodes, and element values ​​are the weighted edge connections (SSW). For example:

[0144] The initial node "Injector carbon deposits" (A) spreads to: the directly adjacent node "Air-fuel ratio imbalance" (B), SSW=0.85; the second-order node "Decrease in combustion efficiency" (C), SSW=0.72; and the third-order node "Excessive nitrogen oxides" (D), SSW=0.53.

[0145] The matrix also labels the edge types (solid lines represent strong causal relationships, and dashed lines represent weak associations), providing a topological basis for subsequent graph updates.

[0146] Based on the weighted edge connections, a time-series graph convolutional network is used to fuse historical health status sequences and update key node scores to obtain a real-time health score matrix.

[0147] Structure design of temporal graph convolutional networks

[0148] The Temporal Graph Convolutional Network (TGCN) consists of three processing layers:

[0149] Spatial Convolutional Layer: Based on the topological relationship of WECM, it aggregates the health status of adjacent nodes. For example, for the node "three-way catalytic converter", it aggregates the features of its upstream "air-fuel ratio" node and downstream "tailpipe emission" node. The aggregation weight is determined by the SSW value in WECM (e.g., the air-fuel ratio node has a weight of 0.7, and the tailpipe node has a weight of 0.6).

[0150] Temporal convolutional layer: A one-dimensional convolutional kernel (kernel length K=5, covering the past 5 working cycles) is used to scan the historical state sequence of the node. For example, for the cylinder pressure stability node, the fluctuation standard deviation sequence of its most recent 5 cycles [0.08, 0.12, 0.15, 0.18, 0.22] is extracted, and the trend feature is output after convolution.

[0151] Gated fusion unit: The ratio of historical information to current update is controlled by the Forget Gate (FG) and Input Gate (IG). For example, when a sudden change in operating condition is detected (such as rapid acceleration), the FG reduces the weight of the historical data to 0.3 and prioritizes the use of the current data.

[0152] Key node scoring update mechanism

[0153] Node scores are calculated based on a multi-factor decay model:

[0154] Base Health Score (BHS): Taken from the initial EHKG baseline value (e.g., piston joint BHS=95 / 100).

[0155] Real-time Attenuation (RTA): Calculated from the characteristic deviation output by TGCN. For example, if the current cylinder wall vibration energy is 20% higher than the reference, then RTA = 15 minutes.

[0156] Propagation Attenuation Factor (PAF): Calculated based on the weighted SSW of the associated edges in WECM. For example, the PAF of the "piston ring wear" node increases by 0.2 times due to its association with the "increased blow-by" node (SSW=0.8).

[0157] The final node score update formula is equivalent to: New score = BHS - RTA × (1 + PAF). For example, for a node with BHS=90, RTA=10, and PAF=0.3, the new score is 90 - 10 × 1.3 = 77.

[0158] Output of real-time health score matrix

[0159] The updated scores from all nodes form a Real-time Health Score Matrix (RHSM). The matrix is ​​partitioned by engine subsystem:

[0160] Combustion domain: includes nodes such as cylinder pressure stability (score 83) and combustion efficiency (76);

[0161] Mechanical domain: piston ring wear (68), valve guide clearance (72), etc.;

[0162] Emission domains: hydrocarbon generation (79), nitrogen oxide pathways (81), etc.

[0163] The matrix is ​​labeled with health level tags (e.g., >80 is green, 60-80 is yellow, <60 is red) for display in the visualization interface.

[0164] Based on the real-time health score matrix, the causal chain of combustion domain-component domain-emission domain is reconstructed using the abnormal path backtracking mechanism to obtain an emission health status map that reflects the current comprehensive emission health status of the engine.

[0165] Triggering conditions for abnormal path backtracking

[0166] The Anomaly Path Backtracking Mechanism (APBM) is triggered in two types of scenarios:

[0167] Explicit trigger: When the score of any node in the RHSM is below the threshold (e.g., emission domain node < 75 points).

[0168] Latent trigger: When the difference in cross-domain correlation scores exceeds the limit (e.g., the combustion domain score is 85 but the emission domain score is only 70, the difference 15 > the preset threshold 10).

[0169] After triggering, the system traces backward along the edges in the WECM from the lowest-scoring node back to the source node with the highest SSW. For example, the backtracking path from "NOx Exceeds Standard" (score 65):

[0170] Excessive nitrogen oxides ← excessively high combustion temperature (SSW=0.8) ← excessively lean air-fuel ratio (SSW=0.9) ← oxygen sensor drift (SSW=0.95)

[0171] Multi-domain causal chain reconstruction technology

[0172] The reconstruction process follows the three-domain coupling rule:

[0173] From the combustion domain to the component domain: verify the consistency of thermodynamic parameters (e.g., a peak in-cylinder temperature of 1800°C requires the corresponding piston top coating to be intact).

[0174] From component area to emission area: Check the compatibility between mechanical condition and chemical reaction chain (e.g., when piston ring clearance > 0.2 mm, unburned hydrocarbons should be > 200 ppm).

[0175] The bidirectional constraint solver eliminates contradictory paths: for example, a backtracking path claims that "the injector blockage causes the air-fuel ratio to be too rich", but the real-time data shows that the exhaust oxygen content (Lambda value = 1.05 > 1.0) is actually too lean, so the path is marked as illegal.

[0176] Generation and updating of emission health status profiles

[0177] The final output Emission Health State Graph (EHSG) consists of a three-layer structure:

[0178] Topology layer: Integrates the updated node scores (e.g., reducing the oxygen sensor node score from 82 to 73) with the verified causal chain (highlighted in red: "oxygen sensor drift → air-fuel ratio imbalance → excessive nitrogen oxides" path).

[0179] Spatiotemporal layer: Embed time-series trends (such as the slope of piston ring wear score over the past 10 minutes -0.8 / minute) in node attributes.

[0180] Decision-making level: Add maintenance recommendations at key nodes (such as associating the "oxygen sensor drift" node with the "recommend cleaning oxygen sensor" instruction).

[0181] The graph ensures real-time performance through an incremental update protocol (IUP): a global update is triggered only when the new score changes by more than Δ=5% compared to the previous one; otherwise, a local refresh is performed.

[0182] The emission health model, built on knowledge graph technology, performs graph attention matching between extracted feature vectors and pre-established knowledge nodes such as combustion theory, failure modes, and emission standards. The health score of each node is dynamically updated through a node state diffusion algorithm, forming a visualized three-dimensional health status topology graph. This enables intelligent mapping from discrete features to system-level health status, intuitively displaying the fault propagation path between engine subsystems, providing a structured reasoning framework for subsequent root cause analysis, and significantly improving the diagnostic efficiency of complex emission problems.

[0183] S204. Based on the emission health status map, perform deviation analysis of map node status and path anomaly detection processing to identify the root cause components or processes that lead to potential or actual emission exceedances, and generate an emission anomaly diagnosis report containing specific fault modes.

[0184] Specifically, based on the emission health status map, the node deviation can be calculated, the KL divergence between the status of key nodes and the health benchmark can be quantified, and the node deviation vector can be obtained.

[0185] Node health benchmark library construction

[0186] The system pre-stores a baseline model of emissions health profiles under the full lifecycle health conditions of the engine. This model is generated through training on a large dataset of tens of thousands of engines of the same model. Each profile node (such as "combustion efficiency," "piston ring sealing," and "NOx generation path") is associated with a multidimensional feature distribution, including a mean vector (MV), a covariance matrix (CM), and a probability density function (PDF). For example, the health baseline for the "combustion efficiency" node is defined as: a peak cylinder pressure waveform of 12.5 ± 0.3 MPa, and a pressure rise rate inflection point located at 8 ± 1 degrees of crankshaft angle (CAD) after top dead center. The baseline library is stored in partitions according to engine operating conditions (such as idle speed, 2000 rpm, and full load) to ensure operating condition matching.

[0187] Real-time node state probabilistic modeling

[0188] In the currently constructed real-time emissions health status map, the status of each node is quantified by a dynamic emission feature vector set (such as the pressure oscillation amplitude of the combustion stability feature vector and the soot concentration of the exhaust gas spectrum). The system extracts the corresponding feature dimension for each node and generates a probability distribution of the real-time status using the kernel density estimation (KDE) algorithm. For example, the features of the "piston ring wear" node include the vibration energy attenuation gradient in the 2-5 kHz frequency band (unit: dB / μm), which is transformed into a probability distribution curve using a Gaussian kernel function and aligned with the probability distribution of the same dimension of the health benchmark.

[0189] KL divergence deviation metric

[0190] The Kullback-Leibler Divergence (KLD) is used as a measure of deviation, and the calculation formula is as follows:

[0191]

[0192] Where P(x) is the real-time state probability distribution and Q(x) is the health baseline probability distribution. For example, when the overlap between the real-time urea injection feedback concentration distribution P(x) and the baseline Q(x) of the "nitrogen oxide generation path" node decreases, the KLD value increases. The system calculates the KLD value in parallel for all key nodes in the map (56 core nodes pre-set) and generates a node deviation vector (NDV). The dimension of this vector is consistent with the number of nodes, and each element represents the degree of abnormality of the corresponding node (unit: information bits). For example, when the KLD of the piston ring wear node is >3.0 bits, a warning threshold is triggered.

[0193] Based on the node deviation vector, the core fault propagation chain leading to excessive emissions is located using the minimum causal path search algorithm;

[0194] Graph Causal Topological Analysis

[0195] The emissions health status graph is essentially a directed weighted graph, where nodes represent states (e.g., "carbon buildup in cylinders," "oxygen sensor failure") and edges represent causal relationships (e.g., "piston ring wear → insufficient cylinder pressure → incomplete combustion"). Each edge is assigned a causal strength weight (CSW), derived from historical failure case statistics and an expert rule base. For example, the CSW from "fuel injector blockage" to "hydrocarbon emissions exceeding limits" is 0.92 (range 0-1), while the CSW from "nitrogen oxide emissions exceeding limits" is only 0.15. The system loads the graph's adjacency matrix (AM) and weight matrix (WM) as the topological basis for path searching.

[0196] Minimal causal path search

[0197] Using nodes whose KLD values ​​exceed the threshold in the node deviation vector as endpoints (e.g., "excessive particulate matter in exhaust gas"), the algorithm searches backwards for possible fault sources. An improved Dijkstra algorithm is employed, with the optimization objective being to maximize the total causal strength of the path (higher causal strength indicates a more likely fault association), while limiting the path length to ≤5 hops (to prevent over-tracing). Starting from the endpoint, the algorithm iteratively traverses incoming neighbor nodes, updates the path's accumulated weight (AW), and records the optimal predecessor node (PN). For example, the optimal path for "excessive particulate matter" is found to be: crankcase ventilation valve blockage (CSW=0.88) → oil vapor backflow (CSW=0.95) → increased carbon deposits in the cylinder (CSW=0.91) → abnormal combustion chamber temperature (CSW=0.89) → excessive particulate matter.

[0198] Core Fault Propagation Chain Generation

[0199] When multiple paths lead to the same source, the path with the highest total path weight (PTW) is selected as the core fault propagation chain (CFPC). Each chain outputs a sequence of nodes, along with local KLD values ​​and path weights. For example, the CFPC for locating excessive nitrogen oxides might be: EGR valve stuck (KLD=4.2bit) → insufficient exhaust gas recirculation rate (KLD=3.8bit) → high in-cylinder temperature (KLD=5.1bit) → thermal nitrogen oxide surge (KLD=6.0bit).

[0200] The system automatically marks critical path nodes and filters out low-confidence chains with PTW < 2.5 (empirical threshold).

[0201] Based on the core fault propagation chain, thermodynamic constraints are verified to eliminate pseudo-paths that violate the temperature-pressure-flow coupling equation, thus obtaining the verified fault chain.

[0202] Engine physical constraint library loading

[0203] The system's pre-built thermodynamic constraint library contains three types of essential physical equations, all of whose parameters are bound to real-time engine sensor data:

[0204] Temperature-pressure coupling equation: based on the actual working fluid state equation

[0205]

[0206] Where P is the combustion chamber pressure in megapascals (MPa); V is the cylinder volume calculated from the piston position in cubic centimeters (cm³); γ is the specific heat ratio of the working fluid, taken as 1.35 for gasoline engines; n is the number of moles of gas in the cylinder, deduced from the intake oxygen concentration; R is the gas constant 8.314; and T1 is the theoretical temperature in Kelvin (K). This equation requires that the real-time cylinder pressure increase must be accompanied by a synchronous increase in the theoretical temperature.

[0207] Flow conservation equation: expressed as in = ex + leak ,in, in Intake mass flow rate, in grams per second (g / s); ex This refers to the exhaust mass flow rate; leakThis represents the crankcase blow-by flow rate. The equation requires that the total amount of gas entering the engine must equal the sum of the exhaust and leakage amounts; an error exceeding 5% will trigger an anomaly.

[0208] Chemical reaction equilibrium equation: describing the rate of formation of nitrogen oxides .in, T1 is the activation temperature for the formation of nitrogen oxides, 38,000 K; T2 is the measured temperature inside the cylinder. The equation represents the percentage of oxygen concentration in the cylinder, and k is a fuel characteristic constant. The equation requires that a high-temperature, high-oxygen environment inevitably leads to an increase in nitrogen oxide concentration.

[0209] For example, a certain CFPC claims that "turbocharger failure leads to insufficient intake pressure → reduced combustion temperature", but if the boost pressure (BP) in the real-time data is 1.8 bar and the peak in-cylinder temperature reaches 1800 Kelvin, then the constraint of "low pressure corresponds to low temperature" is violated.

[0210] Real-time data-driven constraint verification

[0211] For each causal relationship node pair in the core fault propagation chain (e.g., "boost pressure ↓ → combustion temperature ↓"), extract current operating data for verification:

[0212] If there are directly measurable quantities between nodes (such as boost pressure sensor values ​​or in-cylinder thermocouple readings), substitute them into the corresponding equations to calculate the theoretical values ​​and compare them with the measured values ​​to obtain the residual error (RE).

[0213] If there is no direct data, it can be derived through related variables: for example, when verifying "EGR rate ↓ → cylinder temperature ↑", the actual EGR rate is deduced by using the exhaust gas oxygen concentration (EGO) sensor, and then the theoretical cylinder temperature is calculated by combining it with the intake air temperature.

[0214] Set a residual tolerance threshold (e.g., a temperature relative error >15% is considered a violation) and automatically mark node pairs that violate the constraints.

[0215] False path removal and fault chain correction

[0216] For node pairs that violate thermodynamic constraints, remove the edge or the entire path that violates the causal relationship. For example:

[0217] A certain CFPC path is: injector carbon buildup → insufficient fuel injection → reduced cylinder temperature → reduced nitrogen oxides. However, the cylinder temperature actually increases in real-time data (violating the constraint that "insufficient fuel injection should lead to cooling down"), so it is judged to be a false path.

[0218] The corrected version retains the effective sub-chain injector carbon buildup, which leads to poor atomization and increased carbon soot generation.

[0219] The output of the verified path is the Verified Fault Chain (VFC), and each chain is accompanied by a constraint verification pass flag (Verification Flag, VF=1) and residual records.

[0220] Based on the verified fault chain, the fault modes are decoupled to obtain a fault mode list.

[0221] Component-level Failure Mode Mapping

[0222] The Fault Mode Knowledge Base (FMKB) maps graph nodes to specific physical component failure modes:

[0223] Combustion-related issues: such as "combustion oscillation" and corresponding fault modes: injector needle valve sticking, spark plug carbon buildup, cylinder head gasket leakage;

[0224] Emission-related nodes: such as "low nitrogen oxide conversion efficiency" corresponding failure modes: sulfur poisoning of SCR catalyst, urea nozzle blockage;

[0225] Mechanical wear nodes: such as "piston ring axial wear" corresponding to the following failure modes: insufficient oil cleanliness and air filter failure leading to abrasive particle intrusion.

[0226] For example, the "crankcase blow-by flow rate exceeds the standard" node in VFC is decoupled into three possible modes through FMKB: piston ring breakage (probability weight 40%), cylinder liner scratch (35%), and valve guide seal failure (25%).

[0227] Multi-fault chain contention decoupling

[0228] When multiple VFCs point to the same component (e.g., two chains indicating "piston ring wear" and "cylinder liner out-of-roundness" respectively, both leading to excessive blow-by), a Dempster-Shafer Theory (DST) algorithm is used to calculate the belief probability (BP) of each failure mode. The inputs are the path weights (PTW) of each VFC and the node KLD values; the output is a normalized failure probability distribution. For example:

[0229] Piston ring fracture: BP=62% (supporting evidence: blow-by flow + high-frequency vibration attenuation); Cylinder liner scoring: BP=28% (supporting evidence: abnormal temperature gradient + low-frequency vibration); Other: BP=10%.

[0230] Fault mode list generation

[0231] Output a Fault Mode List (FML), sorted in descending order of confidence probability. Each record includes: faulty component (e.g., injector of cylinder 3); specific mode (e.g., needle valve dynamic response delay); confidence probability (e.g., BP=0.78); associated emissions (e.g., hydrocarbon exceedance (HC>50ppm)).

[0232] For example: 1. Faulty component: EGR cooler, mode: cooling pipe blockage, BP=0.85, impact: NOx exceeds the standard; 2. Faulty component: turbocharger, mode: exhaust bypass valve stuck, BP=0.67, impact: particulate matter & CO exceed the standard.

[0233] Based on the failure mode list, component failure probabilities are fused, and combined with real-time data, an emission anomaly diagnostic report with confidence intervals is generated.

[0234] Component-level failure probability calculation

[0235] Based on the confidence probability (BP) in the failure mode list and combined with real-time component stress data, the final failure probability (FP) is calculated:

[0236]

[0237] Where S is the real-time stress (e.g., the offset of piston ring temperature relative to the design value), S_{max} is the maximum allowable stress (e.g., the material's temperature limit), and omega_t is the time decay factor (e.g., the average development rate of the failure mode). For example:

[0238] The failure mode "fuel injector carbon buildup" has a BP of 0.75; the real-time injector temperature is 130°C (design upper limit = 150°C), so S / S_{max} = 0.87; the carbon buildup growth model gives omega_t = 0.92 (based on continuous operating hours); the final FP = 0.75 × 0.87 × 0.92 ≈ 0.60.

[0239] Confidence Interval Construction

[0240] Considering data uncertainty, confidence intervals (CI) are introduced for key parameters:

[0241] Measurement error: For example, a cylinder pressure sensor accuracy of ±0.5MPa will result in CI=[3.2, 4.0] bits for the combustion efficiency node KLD;

[0242] Model error: such as the standard deviation of thermodynamic constraint verification residuals (SD=0.08).

[0243] Probability propagation: The input parameters are randomly perturbed 1000 times using Monte Carlo Simulation (MCS), and the 90% confidence interval of the output FP is obtained (e.g., FP = 0.60 ± 0.07).

[0244] Diagnostic report generation

[0245] Generate a structured emissions anomaly diagnostic report, comprising four parts:

[0246] Root cause summary: such as "EGR cooler blockage (FP=0.85±0.05) caused NOx to exceed the limit to 0.8g / kWh (limit 0.4g / kWh)";

[0247] Fault propagation path: Visualize and verify the fault chain (e.g., EGR rate ↓ → cylinder temperature ↑ → thermal NOx ↑).

[0248] Maintenance priority matrix: sorted by FP and degree of emission exceedance (e.g., immediately handle faults with FP>0.7 and exceedance>50%).

[0249] Real-time data attachment: raw data snippets from key sensors (e.g., coolant temperature difference ΔT=2°C corresponding to a blocked EGR valve → healthy baseline ΔT=10°C);

[0250] The report output is in both PDF and JSON formats, and supports display on in-vehicle terminals and synchronization with the cloud.

[0251] By using KL divergence quantification to measure the deviation of key nodes from the baseline state, and combining it with the minimum causal path search algorithm to locate the fault propagation chain, the system eliminates false correlations through thermodynamic constraint verification. Finally, it outputs a detailed diagnostic report that includes fault component location, failure mechanism description, and confidence assessment. This breaks through the limitations of traditional threshold alarms, enabling fault tracing from phenomenon to essence, accurately identifying the root causes of emission degradation such as injector carbon buildup and EGR valve sticking, and providing a direct basis for precise maintenance.

[0252] S205, based on the emission anomaly diagnosis report and historical emission health status map sequence, apply a prediction algorithm based on map evolution to predict the deterioration trend of specific emission paths of the engine, and generate targeted preventive engine maintenance strategies.

[0253] Specifically, based on the emission anomaly diagnosis report and the historical emission health status map sequence, a spatiotemporal tensor of map evolution can be constructed and node state time slices can be stored to obtain the spatiotemporal tensor of map evolution.

[0254] After the system generates an emission anomaly diagnostic report (including a list of failure modes, confidence intervals, and core failure propagation chains), it needs to combine this report with historical emission health status graph sequences (graph snapshots stored by timestamps) to build a predictive foundation. First, the Graph Evolution Spatiotemporal Tensor (GEST) is a four-dimensional data structure, whose dimensions are defined as follows:

[0255] Dimension 1: Graph Nodes (N). Represent entities in the emissions health graph (such as "piston ring wear", "three-way catalytic converter efficiency", "NOx generation pathway").

[0256] Dimension 2: Node State Features (F). Stores the dynamic state vector of each node (e.g., wear score 0-100, catalytic efficiency percentage, emission concentration ppm).

[0257] Dimension 3: Spatial Hierarchy (H). Layered by engine subsystems (combustion domain, exhaust domain, aftertreatment domain).

[0258] Dimension 4: Time Series (T). Historical map snapshots collected at fixed intervals (e.g., every 10 minutes).

[0259] The construction process begins with the extraction of Node State Time Slice (NSTS): the state vector of each node is extracted from the historical graph sequence in chronological order (for example, the feature value of node "piston ring wear" at time T1 is [wear score=72, vibration energy entropy=0.85]), and the time drift caused by sensor delay is eliminated by the Temporal AlignmentModule (TAM).

[0260] To improve tensor integrity, key information from the emissions anomaly diagnostic report needs to be integrated:

[0261] Fault mode injection: Encode the fault modes in the diagnostic report (such as "piston ring sticking - moderate") into node additional features (such as adding "fault flag bit = 1" and "confidence level = 0.92").

[0262] Causal chain reinforcement: Based on the core fault propagation chain in the report (e.g., "cylinder liner wear → increased blow-by → incomplete combustion → excessive CO"), virtual "causal edge weights" are added to the tensor to help subsequent prediction algorithms focus on critical paths.

[0263] Spatiotemporal correlation imputation: For missing time slices (such as periods of sensor failure), a bidirectional long short-term memory (Bi-LSTM) network is used to complete them. For example, using health map data from one hour before and after the missing time, the node state at the missing time can be predicted.

[0264] Example of the final generated GEST tensor:

[0265] Dimensions: N×F×H×T = 50 nodes × 8-dimensional features × 3 spatial levels × 240 time slices (48 hours of data).

[0266] Stored in a distributed time-series database (such as InfluxDB), it supports millisecond-level query responses, providing structured input for subsequent predictions.

[0267] In the implementation of the project, the construction of GEST needs to solve two major challenges:

[0268] Real-time performance guarantee: Data slicing and alignment are completed on the vehicle-mounted device through edge computing nodes (ECNs), and only the compressed tensors are uploaded to the cloud, reducing bandwidth requirements.

[0269] Scalable design: Hierarchical Tensor Chunking (HTC) is employed to divide large tensors into blocks according to spatial hierarchy (e.g., burning domain subtensors are stored independently) to avoid single-point resource bottlenecks. The output of this step is a spatiotemporally complete GEST tensor, which is the core input for predicting evolution.

[0270] Based on the spatiotemporal tensor of the graph evolution, the spatiotemporal graph neural network is used to predict the state trajectory of key nodes within a preset time period in the future, and the predicted state trajectory is obtained.

[0271] The Spatio-Temporal Graph Neural Network (STGNN) is a prediction engine whose architecture is customized for the GEST tensor.

[0272] Spatial dependency modeling: A graph convolution module (GCM) is used. Based on the connectivity between graph nodes (e.g., the "piston ring wear" node and the "cylinder pressure fluctuation" node have a strong connection), features of adjacent nodes are aggregated. For example, a neighborhood-weighted average of node features is calculated using a Chebyshev polynomial approximation graph convolution kernel.

[0273] Temporal dependency modeling: A gated temporal convolution module (GTCM) is used. Dilated causal convolution (DCC) is employed to capture long-period patterns (such as progressive degradation of wear), and a gating mechanism (similar to GRU) is used to filter noise.

[0274] Spatiotemporal coupling: Spatiotemporal weights are dynamically adjusted through an Attention Fusion Layer (AFL). For example, when a sudden change in emissions is detected, the attention weights in the time dimension are increased for a faster response.

[0275] The prediction process is executed in three stages:

[0276] Key Node Screening: Based on the emissions anomaly diagnostic report, identify high-impact nodes (HINs) that need to be predicted. For example, if the report indicates that "NOx exceedances are caused by EGR valve malfunctions," then EGR valve-related nodes (such as "EGR opening deviation" and "cooling efficiency") are marked as HINs.

[0277] Multi-step rolling prediction: Using the GEST tensor as input, STGNN performs iterative predictions according to a preset time period (e.g., the next 24 hours).

[0278] Input: GEST tensor for the past 48 hours.

[0279] Output: The node state at time t+1 in the future.

[0280] The prediction result at time t+1 is fed back to the input, and the prediction at time t+2 is continued, forming a rolling closed loop.

[0281] Uncertainty quantification: Probabilistic prediction intervals are generated using Monte Carlo Dropout (MCD). For example, the "piston ring wear score" is predicted to be [85, 92] after 24 hours (95% confidence interval).

[0282] The predicted state trajectory (PST) contains two core types of information:

[0283] Node state sequence: Feature values ​​of key nodes at future time points (e.g., predicted values ​​of "three-way catalytic converter efficiency" every 2 hours: [78%, 76%, 74%, ...]).

[0284] Pathway deterioration index: Calculates the rate of deterioration (slope) of a specific emission path (such as the "unburned hydrocarbon generation path"). For example, based on the predicted values ​​of the "air-fuel ratio fluctuation" and "tailpipe HC concentration" nodes, the fitted path deterioration coefficient β = 0.15 (unit: ppm / hour).

[0285] The prediction engine is deployed in a cloud-edge collaborative architecture: the STGNN lightweight model runs on the vehicle ECU, while complex retraining is performed in the cloud to ensure real-time performance.

[0286] Based on the predicted state trajectory, a three-objective Pareto optimization process is performed to find the optimal balance point between emission risk, maintenance cost and downtime, and the optimal intervention plan is obtained.

[0287] Three-objective Pareto optimization (TOPO) requires balancing the following conflicting objectives:

[0288] Objective 1: Emission Risk (ER). Quantify and predict the probability of exceeding standards in the trajectory (e.g., the probability of NOx concentration exceeding the China VI limit of 50 mg / km).

[0289] Objective 2: Maintenance Cost (MC). This includes parts costs (e.g., a piston ring replacement quote of ¥1200) and labor costs (¥300 / hour).

[0290] Objective 3: Downtime Duration (DD). Operational losses caused by vehicle downtime (e.g., a loss of ¥200 per hour for a logistics vehicle).

[0291] The optimization problem is defined as: minimizing [ER, MC, DD]; constraint condition: technical feasibility of maintenance actions (if cylinder block disassembly is required, then DD ≥ 4 hours).

[0292] The solution is obtained using the improved NSGA-III algorithm (a non-dominated sorting genetic algorithm with reference points):

[0293] Genetic Encoding: The maintenance strategy is encoded as a binary gene string. For example, gene bit definitions: Bit 1: Whether to replace the piston rings (0 / 1); Bit 2: Whether to clean the EGR valve (0 / 1); Bit 3: Whether to calibrate the oxygen sensor (0 / 1).

[0294] Fitness assessment:

[0295] ER calculation: If the gene contains "replace piston ring", then the wear node prediction value is updated according to PST, and the probability of exceeding the emission limit is recalculated.

[0296] MC calculation: sum up the part and labor costs for the selected action.

[0297] DD calculation: The maximum value is taken when parallel maintenance actions can overlap (e.g., replacing piston rings takes 3 hours, cleaning EGR valve takes 1 hour, if they are done in parallel, then DD = 3 hours).

[0298] Equilibrium point selection: A Pareto front is generated in the 3D target space, and the most equilibrium solution is selected through entropy-weighted TOPSIS decision-making (Technique for Order Preference by Similarity to Ideal Solution). For example, the solution with the shortest Euclidean distance from the ideal solution (ER=0, MC=0, DD=0) is selected.

[0299] The Optimal Intervention Scheme (OIS) outputs structured instructions:

[0300] Action combination: such as "replace piston rings (priority 1) + calibrate oxygen sensor (priority 2)".

[0301] Execution timing: Set time windows based on the deterioration inflection point of the predicted trajectory (e.g., "piston ring replacement must be performed within 48 hours").

[0302] Economic indicators: Total cost of the plan is ¥1,500, downtime is 2 hours, and emission risk is reduced to 5%.

[0303] Based on the optimal intervention plan, the vehicle operation plan is integrated, and a set of preventive maintenance instructions is output from the database.

[0304] The Vehicle Operation Plan (VOP) is obtained from the fleet management system and includes:

[0305] Itinerary: Transportation routes for the coming week (e.g., "Shanghai → Beijing, 1200km").

[0306] Load spectrum: cargo weight distribution (e.g., outbound trip fully loaded 20 tons, return trip empty).

[0307] Operating condition prediction: Engine load rate based on road conditions (e.g., 85% load rate in mountainous areas).

[0308] The fusion logic is as follows: maintenance actions in the OIS are embedded into the gap window of the VOP, while taking into account the impact of operating conditions on maintenance urgency. For example, if the prediction shows that "piston ring wear deteriorates faster on plateau sections", and the VOP includes transportation on the Qinghai-Tibet Railway, then the priority of this action is automatically increased.

[0309] The process for generating a Preventive Maintenance Instruction Set (PMIS):

[0310] Resource matching: Query the maintenance knowledge graph database:

[0311] Action "Replace Piston Rings" → Associate Required Tools (Cylinder Liner Remover) → Match with Nearby Service Station Inventory.

[0312] Dynamic scheduling: using constraint programming (CP) algorithms:

[0313] Input: VOP time window, service station time pool, spare parts inventory; Output: Optimal repair appointment time (e.g., "June 5th, 10:00, Suzhou service station").

[0314] Instruction layered generation:

[0315] Driver terminal: Pushes "Please schedule maintenance within 48 hours, code P0300"; Maintenance station system: Sends "Work order #202406001: Replace piston ring, 2 hours of work space required"; Supply chain system: Triggers "Piston ring spare parts allocation, destination Suzhou station".

[0316] The instruction set's adaptive mechanism ensures robustness:

[0317] Real-time feedback loop: If the repair station replies "no piston rings in stock," the system automatically triggers alternative solutions:

[0318] Transferred from other sites (increased cost of ¥200, delayed by 1 day); downgraded to "piston ring cleaning + adding engine oil additive" (temporary solution, valid for 72 hours).

[0319] Multi-objective rebalancing: When alternative solutions change MC and DD, TOPO fine-tuning of OIS is re-executed. The final output PMIS includes: a maintenance action list (including spare parts models and man-hours); a time and space scheduling plan (location, time, personnel); an emergency downgrade plan (when the main solution is not feasible); and an economic and environmental benefit assessment (e.g., "expected reduction of NOx emissions by 1.2 kg, avoiding a fine of ¥5,000").

[0320] By learning the evolution patterns of historical health status through spatiotemporal graph neural networks, the system predicts the future trajectory of key parameter changes. Combined with Pareto optimization, it seeks a balance among multiple objectives such as emission risk, maintenance cost, and downtime, and outputs personalized strategies that include the best intervention timing, maintenance content, and expected effects. This enables a shift from passive maintenance to proactive prevention, avoids sudden emission exceedance events through predictive maintenance, optimizes maintenance resource allocation, and significantly reduces the total lifecycle maintenance cost.

[0321] As can be seen, multi-dimensional fusion sensor datasets are collected in real time based on the operating status of the vehicle's internal combustion engine. For these datasets, a dynamic emission status feature vector set is generated, representing the current combustion efficiency, wear status of key components, and emission generation paths. Based on this dynamic emission status feature vector set, an emission health status map reflecting the engine's current overall emission health status is constructed and updated in real time. Based on the emission health status map, an emission anomaly diagnostic report containing specific fault modes is generated. Based on the emission anomaly diagnostic report and historical emission health status map sequences, the deterioration trend of specific emission paths of the engine is predicted, and targeted preventative engine maintenance strategies are generated. This enables real-time and accurate diagnosis and trend prediction of the engine's emission health status, improving the proactiveness and maintenance efficiency of emission anomaly handling.

[0322] Another embodiment of the present invention provides an emission monitoring system for an automotive internal combustion engine, see [link to relevant documentation]. Figure 3 The system may include:

[0323] The acquisition module 301 is used to collect multi-dimensional fusion sensor datasets in real time, including exhaust gas composition spectrum data, combustion chamber pressure fluctuation waveform, cylinder temperature gradient, crankcase blow-by flow rate, and exhaust back pressure, based on the operating status of the automotive internal combustion engine.

[0324] Extraction module 302 is used to perform spatiotemporal alignment and feature extraction on the multidimensional fusion sensing dataset to generate a dynamic emission state feature vector set that characterizes the current combustion efficiency, wear status of key components and emission generation path.

[0325] The construction module 303 is used to map the feature vectors to the map nodes based on the dynamic emission status feature vector set and the pre-constructed emission health map model, and to construct and update the emission health status map that reflects the current comprehensive emission health status of the engine in real time.

[0326] The detection module 304 is used to perform deviation analysis of the status of the nodes in the emission health status map and path anomaly detection processing based on the emission health status map, identify the root cause components or processes that lead to potential or actual emission exceedances, and generate an emission anomaly diagnosis report containing specific fault modes.

[0327] The prediction module 305 is used to predict the deterioration trend of a specific emission path of the engine based on the emission anomaly diagnosis report and the historical emission health status map sequence, and to generate targeted preventive engine maintenance strategies.

[0328] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0329] Specifically, in this embodiment, the storage medium can be configured to store a computer program for performing the following steps:

[0330] S201 collects multi-dimensional fusion sensor datasets in real time, including exhaust gas composition spectrum data, combustion chamber pressure fluctuation waveform, cylinder temperature gradient, crankcase blow-by flow rate, and exhaust back pressure, based on the operating status of the automotive internal combustion engine.

[0331] S202, For the multi-dimensional fusion sensing dataset, perform spatiotemporal alignment and feature extraction to generate a dynamic emission state feature vector set that characterizes the current combustion efficiency, wear status of key components and emission generation path;

[0332] S203, Based on the dynamic emission status feature vector set, the feature vectors are mapped to the map nodes using the pre-constructed emission health map model, and an emission health map reflecting the current comprehensive emission health status of the engine is constructed and updated in real time.

[0333] S204. Based on the emission health status map, perform deviation analysis of map node status and path anomaly detection processing to identify the root cause components or processes that lead to potential or actual emission exceedances, and generate an emission anomaly diagnosis report containing specific fault modes.

[0334] S205, based on the emission anomaly diagnosis report and historical emission health status map sequence, apply a prediction algorithm based on map evolution to predict the deterioration trend of specific emission paths of the engine, and generate targeted preventive engine maintenance strategies.

[0335] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0336] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0337] Specifically, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0338] S201 collects multi-dimensional fusion sensor datasets in real time, including exhaust gas composition spectrum data, combustion chamber pressure fluctuation waveform, cylinder temperature gradient, crankcase blow-by flow rate, and exhaust back pressure, based on the operating status of the automotive internal combustion engine.

[0339] S202, For the multi-dimensional fusion sensing dataset, perform spatiotemporal alignment and feature extraction to generate a dynamic emission state feature vector set that characterizes the current combustion efficiency, wear status of key components and emission generation path;

[0340] S203, Based on the dynamic emission status feature vector set, the feature vectors are mapped to the map nodes using the pre-constructed emission health map model, and an emission health map reflecting the current comprehensive emission health status of the engine is constructed and updated in real time.

[0341] S204. Based on the emission health status map, perform deviation analysis of map node status and path anomaly detection processing to identify the root cause components or processes that lead to potential or actual emission exceedances, and generate an emission anomaly diagnosis report containing specific fault modes.

[0342] S205, based on the emission anomaly diagnosis report and historical emission health status map sequence, apply a prediction algorithm based on map evolution to predict the deterioration trend of specific emission paths of the engine, and generate targeted preventive engine maintenance strategies.

[0343] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A method of monitoring the exhaust emissions of an internal combustion engine of an automotive vehicle, characterised in that, The method comprises: According to the operating state of the automobile internal combustion engine, a multi-dimensional fusion sensing data set including tail gas component spectral data, combustion chamber pressure fluctuation waveform, in-cylinder temperature gradient, crankcase blow-by flow and exhaust back pressure is synchronously collected in real time; For the multi-dimensional fusion sensing data set, time-space alignment and feature extraction are performed to generate a dynamic emission state feature vector set representing the current combustion efficiency, key component wear state and emission generation path; According to the dynamic emission state feature vector set, the emission health atlas model is pre-constructed, the feature vector is mapped to the atlas node, and the emission health state atlas reflecting the current comprehensive emission health status of the engine is constructed and updated in real time; According to the emission health state atlas, the atlas node state deviation degree analysis and path anomaly detection processing are performed to identify the root cause components or processes leading to potential or actual emission exceedance, and an emission anomaly diagnosis report containing specific fault modes is generated; According to the emission anomaly diagnosis report and the historical emission health state atlas sequence, a prediction algorithm based on atlas evolution is applied to predict the deterioration trend of the specific emission path of the engine, and a targeted preventive engine maintenance strategy is generated.

2. The method of claim 1, wherein, According to the operating state of the automobile internal combustion engine, a multi-dimensional fusion sensing data set including tail gas component spectral data, combustion chamber pressure fluctuation waveform, in-cylinder temperature gradient, crankcase blow-by flow and exhaust back pressure is synchronously collected in real time; According to the phase synchronization pulse signal output by the crankshaft position sensor, the sampling clock of the cylinder pressure sensor, the exhaust gas spectrometer, the temperature array, the blow-by flow meter and the back pressure sensor is synchronized to obtain time reference aligned original signal stream; According to the time reference aligned original signal stream, combustion cycle segmentation is performed, and the compression stroke and work stroke data slices are divided according to the crank angle to obtain a combustion stage grouping signal set; According to the combustion stage grouping signal set, engine geometric space mapping is performed, and the cylinder wall vibration signal is mapped to the piston ring-cylinder liner friction pair position based on piston kinematics to obtain a space registration vibration feature matrix; According to the space registration vibration feature matrix, multi-source noise cancellation processing is performed to fuse the exhaust gas spectrum, temperature and flow data and suppress ignition electromagnetic interference to obtain a time-space synchronous multi-dimensional fusion sensing data set.

3. The method of claim 2, wherein, The method comprises: According to the multi-dimensional fusion sensing data set, cylinder pressure waveform variational modal decomposition is performed to extract pressure rise rate inflection point and indicated mean effective pressure energy distribution features to obtain a combustion stability feature vector; According to the exhaust gas spectral data, chemical bond absorption peaks are associated, and emission concentration proportion and conversion path are analyzed to obtain a pollutant generation path feature vector; According to the space registration vibration feature matrix, the harmonic wavelet packet entropy algorithm is used to quantify the 2-5kHz frequency band energy attenuation gradient to obtain a piston ring wear state feature vector; The combustion stability feature vector, the pollutant generation path feature vector and the piston ring wear state feature vector are fused to construct a three-dimensional dynamic emission state tensor; According to the three-dimensional dynamic emission state tensor, non-negative constraint tensor decomposition is performed to extract a low-rank feature subspace representing emission degradation, thereby obtaining a dynamic emission state feature vector set.

4. The method of claim 3, wherein, According to the dynamic emission state feature vector set, a pre-constructed emission health atlas model is used to map the feature vector to an atlas node, and an emission health state atlas reflecting the current comprehensive emission health status of the engine is constructed and updated in real time, including: According to the dynamic emission state feature vector set, a graph attention network matching is performed to map the nearest neighbor node in the pre-constructed emission health knowledge graph, thereby obtaining an initial node mapping relationship; According to the initial node mapping relationship, a node state diffusion algorithm is used to calculate the semantic similarity weight of the dynamic emission state feature vector and the node, thereby obtaining a weighted edge connection relationship; According to the weighted edge connection relationship, a time series graph convolution network is used to fuse the historical health state sequence and update the key node score, thereby obtaining a real-time health score matrix; According to the real-time health score matrix, an abnormal path backtracking mechanism is used to reconstruct the combustion domain-component domain-emission domain causal chain, thereby obtaining an emission health state atlas reflecting the current comprehensive emission health status of the engine.

5. The method of claim 4, wherein, According to the emission health state atlas, a graph node state deviation degree analysis and path anomaly detection process are performed to identify the root cause components or processes leading to potential or actual emission exceedance, and an emission anomaly diagnosis report containing specific fault modes is generated, including: According to the emission health state atlas, a node deviation degree calculation is performed to quantify the KL divergence of the key node state and the health benchmark, thereby obtaining a node deviation degree vector; According to the node deviation degree vector, a minimum causal path search algorithm is used to locate the core fault propagation chain causing the emission exceedance; According to the core fault propagation chain, a thermodynamic constraint verification is performed to filter out false paths that violate the temperature-pressure-flow coupling equation, thereby obtaining a verified fault chain; According to the verified fault chain, a fault mode decoupling is performed to obtain a fault mode list; According to the fault mode list, a component failure probability fusion is performed to generate an emission anomaly diagnosis report with a confidence interval based on real-time data.

6. The method of claim 5, wherein, According to the emission anomaly diagnosis report and the historical emission health state atlas sequence, a prediction algorithm based on atlas evolution is applied to predict the deterioration trend of a specific emission path of the engine, and a targeted preventive engine maintenance strategy is generated, including: According to the emission anomaly diagnosis report and the historical emission health state atlas sequence, a graph evolution spatiotemporal tensor is constructed, and a node state time slice is stored, thereby obtaining a graph evolution spatiotemporal tensor; According to the graph evolution spatiotemporal tensor, a spatiotemporal graph neural network is used to predict the key node state trajectory in the future preset period, thereby obtaining a predicted state trajectory; According to the predicted state trajectory, a three-objective Pareto optimization process is performed to solve the optimal balance point of emission risk, maintenance cost and downtime length, thereby obtaining an optimal intervention scheme; According to the optimal intervention scheme, a preventive maintenance instruction set is output by combining the vehicle operation plan and the database.

7. An exhaust emission monitoring system for an internal combustion engine of an automobile, characterized by comprising: The system comprises: The acquisition module is configured to acquire, in real time and synchronously, a multi-dimensional fusion sensing data set including tail gas component spectral data, combustion chamber pressure fluctuation waveform, in-cylinder temperature gradient, crankcase blow-by flow and exhaust back pressure according to the operating state of the internal combustion engine of the automobile; The extraction module is configured to perform time-space alignment and feature extraction on the multi-dimensional fusion sensing data set to generate a dynamic emission state feature vector set representing current combustion efficiency, key component wear state and emission generation path; The construction module is configured to map the feature vector to a graph node by using a pre-constructed emission health graph model according to the dynamic emission state feature vector set, and construct and update an emission health state graph reflecting the current comprehensive emission health condition of the engine in real time; The detection module is configured to perform graph node state deviation analysis and path anomaly detection processing according to the emission health state graph, identify the root cause components or processes leading to potential or actual emission exceedance, and generate an emission anomaly diagnosis report containing specific fault modes; The prediction module is configured to apply a graph evolution-based prediction algorithm to predict the deterioration trend of a specific emission path of the engine according to the emission anomaly diagnosis report and a historical emission health state graph sequence, and generate a targeted preventive engine maintenance strategy.

8. The system of claim 7, wherein, The acquisition module is specifically configured to: Synchronize the sampling clocks of the cylinder pressure sensor, the tail gas spectrometer, the temperature array, the blow-by flow meter and the back pressure sensor according to the phase synchronization pulse signal output by the crankshaft position sensor to obtain time reference aligned original signal streams; Divide the combustion cycle according to the time reference aligned original signal streams, divide the compression stroke and work stroke data slices according to the crank angle, and obtain a combustion stage grouped signal set; Map the engine geometric space according to the combustion stage grouped signal set, map the cylinder wall vibration signal to the piston ring-cylinder liner friction pair position based on the piston kinematics, and obtain a space registration vibration feature matrix; Perform multi-source noise cancellation processing according to the space registration vibration feature matrix, fuse the tail gas spectrum, temperature and flow data, and suppress the ignition electromagnetic interference to obtain a time-space synchronous multi-dimensional fusion sensing data set.

9. A storage medium, characterized by The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-6 when running.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the computer program to execute the method of any one of claims 1-6.

Images