Engine abnormality detection method using gas sensor combined with engineering mechanics model

Abstract

The present application relates to the technical field of engine testing and fault diagnosis, and particularly relates to an engine abnormality detection method combining a gas sensor and an engineering mechanics model. The method aims to solve the problem that early abnormal wear of key friction pairs of an engine is difficult to realize online and accurate positioning and has a high false alarm rate. The method comprises the following steps: synchronously acquiring an instantaneous speed fluctuation signal and exhaust composition data of the engine; performing frequency domain analysis on the instantaneous speed fluctuation signal to separate out a low-frequency characteristic signal reflecting friction loss; using a pre-established engine tribology system model to output a theoretical prediction mode for wear of a specific friction pair; matching the actual characteristic signal with the theoretical prediction mode to obtain a matching degree, and analyzing the change trend of carbon monoxide in the exhaust composition; finally, only when the mode matching degree for a certain friction pair exceeds a threshold value and it is simultaneously determined that there is an abnormal trend of the exhaust composition, early wear warning information for the friction pair is generated.

Description

Technical Field

[0001] This invention relates to the field of engine testing and fault diagnosis technology. Specifically, it relates to an engine anomaly detection method that combines a gas-sensitive sensor with an engineering mechanics model, which is particularly suitable for non-invasive online early warning and location of early abnormal wear of key friction pairs in engines. Background Technology

[0002] The engine is the core of vehicles, ships, and various power equipment. Components such as piston rings and cylinder liners, and crankshaft journals and bearings, directly determine the engine's reliability, efficiency, and lifespan. Abnormal wear of these friction pairs is a gradual process. In the early stages of wear, the trace metal particles generated mainly remain in the engine oil and rarely enter the exhaust passage.

[0003] Currently, there are two main technical approaches for monitoring this type of wear: one is offline oil analysis technology, such as ferrography. Although this method can accurately detect wear particles, it has inherent drawbacks such as sampling lag, the need for machine shutdown, and the inability to achieve continuous online monitoring; the other is vibration signal analysis technology. Although this method can be implemented online, it is not sensitive to the weak signals caused by early wear, and the vibration signals are easily interfered with by various vibration sources such as combustion impact and mechanical background noise, making it difficult to accurately distinguish the specific location of wear, that is, it is difficult to achieve precise "location" of the fault source.

[0004] In existing technologies, there are also some methods that attempt to monitor the mechanical condition of an engine online. For example, Chinese patent document CN117738764A discloses a "method and vehicle for monitoring wear of engine friction pairs". This method mainly monitors the vehicle's vibration acceleration in the pure electric driving mode of a hybrid vehicle and triggers active lubrication based on the accumulated vibration damage, aiming to prevent the occurrence or aggravation of wear. This technology is preventative protection, and its technical means rely on the threshold judgment of a single vibration signal, without involving early diagnosis, type identification, and precise location of wear that has already occurred. When wear has already occurred but has not yet caused the overall vibration energy to significantly exceed the standard, such methods are difficult to detect, let alone determine which specific friction pair the wear originates from.

[0005] Therefore, existing technologies lack an effective means to provide early warning and location of abnormal wear of specific friction pairs in a non-intrusive, real-time manner while the engine is operating normally. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an engine anomaly detection method combining a gas-sensitive sensor and an engineering mechanics model. This method aims to solve the technical problem of how to achieve online, real-time, and precise location and early warning of early abnormal wear of key internal friction pairs without disassembling the engine. To achieve the above objective, the embodiments of this invention disclose the following technical solutions:

[0007] A method for detecting engine anomalies using a gas-sensitive sensor combined with an engineering mechanics model, the method being executed by an onboard processor, comprising:

[0008] S110, acquire the instantaneous speed fluctuation signal and exhaust composition data of the engine under the current operating conditions, wherein the instantaneous speed fluctuation signal is collected by a high-precision crankshaft speed sensor and reflects the instantaneous angular velocity change of the flywheel gear, and the exhaust composition data includes at least the carbon monoxide concentration collected by a gas-sensitive sensor;

[0009] S120, perform frequency domain analysis on the instantaneous speed fluctuation signal to separate the low-frequency component that does not overlap with the combustion excitation frequency band, thereby obtaining the actual friction loss characteristic signal that reflects the changes in reciprocating inertial force and friction torque;

[0010] S130, Input the operating condition parameters of the current operating condition into the pre-established engine tribological system model, and run the model to output the predicted friction torque change pattern when at least one preset friction pair experiences a preset degree of wear under the current operating condition, wherein the engine tribological system model is configured to simulate the friction torque change characteristics of different friction pairs under different degrees of wear.

[0011] S140, perform pattern matching analysis on the actual friction loss characteristic signal and the predicted friction torque change pattern to calculate the pattern matching degree for the preset friction pair. At the same time, based on the carbon monoxide concentration change trend, determine whether there is an abnormal trend of exhaust components associated with the decrease in mechanical efficiency.

[0012] S150, if and only if the mode matching degree for a certain preset friction pair exceeds a preset matching threshold, and an abnormal trend of the exhaust composition is simultaneously determined, an early abnormal wear warning message for the preset friction pair is generated and output, the warning message containing the identifier of the friction pair being warned.

[0013] Optionally, the frequency domain analysis in S120 may specifically include: performing a fast Fourier transform on the instantaneous speed fluctuation signal to obtain a frequency spectrum; identifying the combustion excitation frequency band corresponding to the engine ignition frequency and its harmonics in the frequency spectrum; extracting the spectral components in the frequency spectrum that are lower than the lowest frequency of the combustion excitation frequency band and performing an inverse transform to obtain the actual friction loss characteristic signal.

[0014] Optionally, the method for calculating the pattern matching degree in S140 is to calculate the cross-correlation coefficient, specifically using the following formula:

[0015]

[0016] in, The first characteristic signal of actual friction loss Data from each sampling point, The first of the predicted friction torque variation patterns Data from each sampling point, and These are their means, The total number of sampling points. That is, the cross-correlation number. The numerical range of is [-1, 1], and the closer its absolute value is to 1, the higher the matching degree.

[0017] Optionally, the "determining whether there is an abnormal trend in exhaust components associated with a decrease in mechanical efficiency" in S140 is specifically achieved through the following steps: continuously monitoring the time-varying sequence data of carbon monoxide concentration under stable operating conditions; performing linear trend fitting on the time-varying sequence data of carbon monoxide concentration to calculate the average upward slope of its concentration over time; comparing the calculated average upward slope with a first threshold obtained by querying a preset dynamic slope threshold table based on the current operating conditions; if the average upward slope is greater than or equal to the first threshold, it is determined that there is an abnormal trend in exhaust components; wherein, the dynamic slope threshold table stores verified and determined slope thresholds corresponding to different engine speed and load combinations.

[0018] Optionally, the "pre-established engine tribological system model" is obtained through the following offline construction steps: establishing a high-fidelity simulation model that includes the multibody dynamics of the engine crankshaft and connecting rod mechanism and the hydrodynamic lubrication theory of key friction pairs; in the simulation model, defining an independent and continuously adjustable wear state parameter for each preset friction pair to be monitored, wherein the wear state parameter is used to quantitatively characterize the increase in the equivalent lubrication clearance or the increase in the equivalent surface roughness of the friction pair; systematically adjusting the wear state parameter of each preset friction pair under multiple discrete operating conditions covering the typical operating range of the engine, running the simulation model and recording the output total friction torque fluctuation time series data; organizing all simulation results to construct a mapping relationship database with the combination of operating condition parameters and wear state parameters as input and the corresponding friction torque fluctuation characteristic pattern as output.

[0019] Optionally, the "generating and outputting early abnormal wear warning information for the preset friction pair" in S150 further includes a step of quantifying the severity of wear: mapping the pattern matching degree to one of a preset severity level range according to the specific value of the pattern matching degree exceeding the matching threshold; in addition to the friction pair identifier, the warning information also includes a data field for recording the determined severity level, thereby forming a composite warning information that includes the fault location and the preliminary quantified severity.

[0020] Optionally, before executing S110, a pre-screening step is also included: real-time monitoring of the engine's speed signal and load signal, and calculation of their fluctuation within a preset time window; only when the speed fluctuation falls within a first preset range and the load fluctuation falls within a second preset range is it determined that the engine has entered a stable diagnostic operating condition window that can be used for accurate diagnosis, and S110 to S150 are triggered; if the engine has not entered the stable diagnostic operating condition window, the diagnostic process is paused or only data caching and recording are performed.

[0021] Optionally, the "determination of whether there is an abnormal trend in exhaust composition associated with a decrease in mechanical efficiency" further includes a supplementary verification step: while making a judgment based on the carbon monoxide concentration trend, the real-time air-fuel ratio of the engine is also calculated based on the oxygen concentration data in the exhaust composition data; the calculated real-time air-fuel ratio is compared with the normal theoretical range of air-fuel ratio determined based on the current operating conditions and engine calibration data; if the real-time air-fuel ratio continues to deviate from the normal theoretical range and reaches a preset deviation tolerance, this is used as auxiliary evidence to strengthen the conclusion of the "abnormal trend in exhaust composition".

[0022] Optionally, after S150, a closed-loop control step is further included: the generated early abnormal wear warning information is sent to the engine control unit via the vehicle communication network; the engine control unit is pre-programmed to adaptively execute a predefined protective control strategy based on the friction pair identifier and severity level contained in the warning information after receiving it. The protective control strategy includes at least one of the following: specifically limiting the torque output of a specific cylinder, increasing the oil pressure setpoint of the main oil passage or a specific lubrication branch, or activating a pulse-type enhanced lubrication mode for suspected wear areas.

[0023] Optionally, the method further includes a long-term health status prediction step: establishing a historical database for persistently storing the warning information generated during each execution of S150, the corresponding pattern matching degree values, and the operating condition parameters at that time; based on the historical database, performing regression analysis on the change sequence of the pattern matching degree values ​​of the specified friction pair over time or the cumulative working mileage of the engine, and calculating its long-term trend slope; if the analysis finds that the long-term trend slope exceeds a preset trend alarm threshold, generating and outputting a predictive maintenance prompt message, which indicates that the performance degradation of the corresponding friction pair is accelerating and suggests arranging priority inspection.

[0024] The beneficial effects of this invention are as follows:

[0025] (1) By using existing or easily installed standard sensors in the engine, continuous online monitoring of the health status of internal friction pairs can be achieved without disassembling the engine, enabling early, online and non-invasive diagnosis.

[0026] (2) By performing high-precision pattern matching between the processed actual signal and the theoretical signal predicted by the engine tribological system model for a specific preset friction pair under different wear levels, the source of the anomaly can be locked to the specific friction pair, thus overcoming the problem of accurate fault location.

[0027] (3) Through the dual-path collaborative judgment mechanism of "mechanical signal matching" and "chemical trend verification", false positives caused by interference from a single signal can be effectively suppressed, making the diagnostic conclusion more reliable.

[0028] (4) It can not only output location and quantitative diagnostic warnings, but also transmit information to the control unit through the vehicle network to trigger adaptive protective control strategies; at the same time, by tracking and analyzing wear trends through long-term historical data, it can achieve the sublimation from immediate fault warning to predictive life assessment. Attached Figure Description

[0029] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0030] Figure 1 The following is a general flowchart of the engine anomaly detection method combining a gas sensor and an engineering mechanics model provided in an embodiment of the present invention;

[0031] Figure 2This is a flowchart illustrating the signal processing for separating the actual friction loss characteristic signal from the instantaneous speed fluctuation signal in an embodiment of the present invention.

[0032] Figure 3 This is a schematic diagram illustrating the principle of constructing and running the engine tribological system model in an embodiment of the present invention;

[0033] Figure 4 This is a logical diagram illustrating the steps for integrating judgment and early warning information generation in an embodiment of the present invention. Detailed Implementation

[0034] Specific embodiments of the invention will now be described in detail. Although the invention is described in conjunction with these specific embodiments, it should be understood that it is not intended to limit the invention to these specific embodiments. Rather, these embodiments are intended to cover alternative, modified, or equivalent embodiments that may be included within the spirit and scope of the invention as defined by the claims. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. The invention may be practiced without some or all of these specific details.

[0035] When used in conjunction with the terms "comprising," "method comprising," or similar language in this specification and appended claims, the singular forms "a," "some," and "the" include plural references unless the context clearly indicates otherwise. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] The following description, with reference to the accompanying drawings, illustrates an engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model, according to an embodiment of the present invention. Addressing the aforementioned background art's goal of achieving online, real-time, and precise location and early warning of early abnormal wear of key internal friction pairs without disassembling the engine, and significantly improving the reliability of diagnostic conclusions to reduce false alarms, the engine anomaly detection method of the present invention, employing a gas-sensitive sensor and an engineering mechanics model, fuses mechanical dynamics signals with combustion chemical reaction information, and utilizes a high-fidelity tribological model for intelligent analysis and collaborative verification, thereby achieving online, precise location, and highly reliable early warning of early wear of specific friction pairs within the engine.

[0037] Specifically, this method fuses two independent information sources—mechanical (friction characteristic signals analyzed based on speed fluctuations) and chemical (efficiency trends based on exhaust composition changes)—forming a "logical AND" decision-making mechanism. This ensures that the system only issues a warning when both pieces of evidence are simultaneously supported, effectively suppressing false positives caused by random interference to a single signal (such as road impacts or instantaneous sensor noise). This results in a diagnostic confidence level far exceeding that of traditional methods relying on a single signal source. Through a pre-established engine tribological system model based on multibody dynamics and fluid lubrication theory, this method can generate "theoretical fault fingerprints" (i.e., predicted friction torque change patterns) for different preset friction pairs under varying wear levels. High-precision pattern matching between measured signals and these theoretical fingerprints allows the system to pinpoint the anomaly source to a specific friction pair (such as the "piston ring assembly of the first cylinder"), rather than issuing a general "engine mechanical wear" warning, significantly enhancing the decision-making and maintenance guidance value of the diagnostic information. When judging abnormal trends in exhaust composition, this method introduces a slope threshold that dynamically adjusts with real-time engine operating conditions (speed, load). This mechanism fully considers the normal fluctuation range of exhaust components under different operating conditions, avoiding the problem of being overly sensitive or sluggish when using fixed thresholds under certain operating conditions. This allows for more reliable detection of subtle early trends indicating a systematic decline in mechanical efficiency in complex real-world operating environments, improving detection sensitivity and practicality. This method not only outputs locational and quantitative diagnostic warnings but also transmits information to the control unit via the vehicle network, triggering adaptive protective control strategies (such as torque limiting and enhanced lubrication). Simultaneously, by tracking and analyzing wear trends using long-term historical data, it enables a shift from immediate fault warnings to predictive life assessments. This forms a complete closed-loop health management system integrating state perception, real-time diagnosis, active protection, and trend prediction, laying a solid technical foundation for predictive engine maintenance.

[0038] The following is in conjunction with the appendix Figure 1-4 The preferred embodiments of the present invention will be described in detail so that those skilled in the art can implement the present invention accordingly.

[0039] A method for detecting engine anomalies using a gas-sensitive sensor combined with an engineering mechanics model, comprising:

[0040] S110 acquires the instantaneous speed fluctuation signal and exhaust composition data of the engine under the current operating conditions. The instantaneous speed fluctuation signal is collected by a high-precision crankshaft speed sensor and reflects the instantaneous angular velocity change of the flywheel gear. The exhaust composition data includes at least the carbon monoxide concentration collected by a gas-sensitive sensor.

[0041] It is understood that this step is the starting point for data input and synchronous acquisition stage of the engine anomaly detection method that combines gas sensors with engineering mechanical models. Its purpose is to provide two types of synchronous raw signals reflecting the engine's mechanical and thermochemical states for subsequent fusion diagnosis.

[0042] The acquisition of instantaneous speed fluctuation signals relies on a high-precision crankshaft speed sensor. This sensor is typically mounted on the engine block, maintaining a specific gap with the engine flywheel sprocket. Its "high precision" is reflected in its ability to detect the passage time of each tooth (or a specific marked tooth) on the flywheel at a resolution far exceeding the crankshaft's fundamental rotational frequency. For example, for a 60-tooth flywheel, at an engine speed of 3000 rpm (i.e., 50 rpm), its fundamental tooth passage frequency is 3000 Hz. The high-precision sensor can accurately measure the passage time of each tooth, and then calculate the time difference between consecutive teeth. The instantaneous angular velocity of the flywheel can be deduced from this. ( These tiny sequences of instantaneous angular velocity changes constitute the "instantaneous speed fluctuation signal." This signal not only contains fluctuations caused by uneven work done by periodic combustion in each cylinder, but also contains, at a deeper level, information on torsional vibrations excited by the reciprocating inertial force and the changing resistance torque of each friction pair (such as piston ring assembly and crankshaft bearing). It is the original carrier for extracting the characteristics of mechanical friction loss.

[0043] The acquisition of exhaust composition data includes at least the carbon monoxide concentration collected by a gas sensor installed in the engine exhaust manifold or at a suitable downstream location. Carbon monoxide is a product of incomplete combustion, and its concentration is extremely sensitive to the engine's air-fuel ratio and combustion efficiency. Monitoring carbon monoxide is chosen because when engine mechanical efficiency decreases due to increased friction, the engine control system may tend to fine-tune the fuel injection strategy to maintain the same output power, or the combustion conditions themselves may experience subtle deterioration. These often lead to observable trends in carbon monoxide emissions. Therefore, the trend in carbon monoxide concentration can serve as a key chemical indicator reflecting whether there is a systematic decline in the overall thermodynamic efficiency of the engine. In practical implementation, the gas sensor can be based on the non-dispersive infrared principle to ensure measurement accuracy and long-term stability.

[0044] For effective fusion analysis, instantaneous engine speed fluctuation signals and exhaust component data must be acquired synchronously or with a defined time correlation. This is typically achieved through a unified timing clock or controller area network (CLAN) synchronization signal for the onboard processor, ensuring that subsequent steps analyze data based on the same time period or the same engine operating cycle. Furthermore, the acquired raw data usually undergoes preliminary preprocessing, such as analog-to-digital conversion, resampling at a fixed sampling rate, and necessary filtering to eliminate some circuit noise, preparing it for the next stage of sophisticated signal processing.

[0045] Acquiring these two seemingly unrelated signals simultaneously lays the data foundation for subsequent cross-domain information fusion analysis and accurate diagnosis of friction pairs, which is a prerequisite for achieving non-invasive intelligent diagnosis.

[0046] It should be noted that before executing S110, there is a pre-screening step: real-time monitoring of the engine speed and load signals, and calculation of their fluctuation within a preset time window; only when the speed fluctuation falls within the first preset range and the load fluctuation falls within the second preset range is the engine determined to have entered a stable diagnostic operating condition window that can be used for accurate diagnosis, and S110 to S150 are triggered; if the stable diagnostic operating condition window is not entered, the diagnostic process is paused or only data caching and recording are performed.

[0047] Specifically, the main function of the pre-screening step is to determine whether the engine is in a stable operating state suitable for analysis before performing precise diagnostics. This is because, under transient conditions such as rapid acceleration and deceleration, the intake, fuel injection, and combustion processes fluctuate drastically, resulting in a large number of non-stationary interference components in the engine speed and exhaust signals. These interferences can severely mask the weak characteristic signals caused by friction pair wear, thus affecting the accuracy of subsequent analysis. This step, as a feedforward control element, can effectively improve the robustness and reliability of the entire diagnostic system.

[0048] "Real-time monitoring of engine speed and load signals" refers to continuously reading parameters from the engine control network (such as the CAN bus) or direct sensors. The speed signal typically originates from a regular pulse sequence from the crankshaft position sensor. The load signal is a parameter that comprehensively characterizes the engine's output power demand. In practice, it can be equivalently represented by the load rate (expressed as a percentage) calculated by the engine control unit, intake manifold absolute pressure (MAP), throttle opening, or fuel injection quantity per cycle. Choosing a signal that directly reflects the engine's output energy as the load indicator is crucial. "Calculating its fluctuation within a preset time window" refers to a quantitative evaluation of the stability of the two signal sequences. The length of the preset time window needs to balance response speed and statistical stability; for example, it can select data segments from the most recent 2 seconds or N consecutive engine working cycles. The method for calculating the fluctuation must effectively reflect the signal's stability, typically using standard deviation, range, coefficient of variation, etc. The "first preset range" and "second preset range" are dynamic thresholds for determining whether a steady state exists. These ranges are not fixed values ​​but rather empirical boundaries determined through engine calibration tests. During the calibration process, under various steady-state operating conditions (such as different speed and load combinations), the normal distribution range of the fluctuation of the healthy engine speed and load signal is statistically analyzed (for example, taking the 95% confidence interval), and an appropriate margin is left on this basis to form the preset range. For example, for a gasoline engine at a steady-state point of 2000 rpm and 50% load, the first preset range of its speed fluctuation may be calibrated as ±15 rpm (i.e., the fluctuation must be less than 15 rpm), and the second preset range of its load fluctuation may be calibrated as ±3%.

[0049] The condition that "the engine is only considered to have entered the stable diagnostic condition window when the speed fluctuation falls within the first preset range and the load fluctuation falls within the second preset range" constitutes a logical AND condition. Dual criteria are more stringent and reliable than single criteria. Only when the engine speed and load remain stable at the same time can its operating state be considered sufficient to support the extraction of consistent friction characteristics and exhaust trends, thereby triggering the subsequent diagnostic process starting from S110.

[0050] If the above dual conditions are not met—that is, the engine is undergoing transient processes such as rapid acceleration, rapid deceleration, or cold start warm-up—the system will suspend the computationally intensive feature extraction, model matching, and fusion judgment (S120-S150) to conserve computing resources and prevent false alarms. Simultaneously, the system can choose to enter "data cache recording" mode, storing raw data or simply processed data streams in a cache area according to timestamps. This cached data can be used for subsequent offline analysis, operating condition statistics, or, after the vehicle enters a stable window, provide continuous background data for historical trend analysis, thus ensuring no potentially useful information is lost.

[0051] S120 performs frequency domain analysis on the instantaneous speed fluctuation signal to separate the low-frequency components that do not overlap with the combustion excitation frequency band, thereby obtaining the actual friction loss characteristic signal that reflects the changes in reciprocating inertial force and friction torque.

[0052] The frequency domain analysis specifically includes: performing a fast Fourier transform on the instantaneous speed fluctuation signal to obtain the frequency spectrum; identifying the combustion excitation frequency band corresponding to the engine ignition frequency and its harmonics in the frequency spectrum; extracting the spectral components in the frequency spectrum that are lower than the lowest frequency of the combustion excitation frequency band and performing an inverse transform, thereby obtaining the actual friction loss characteristic signal that characterizes the changes in reciprocating inertial force and friction torque.

[0053] Specifically, this step aims to selectively extract the components primarily contributed by the friction pair state from the complex raw speed signal. Engine speed fluctuations are the result of multiple excitation sources superimposed, among which the components with the largest amplitude and highest frequency are usually directly related to the periodic combustion pressure impact (combustion excitation) of the cylinder. The torsional vibration response excited by changes in the friction pair state typically has a frequency response significantly lower than the dominant frequency of the combustion excitation. Therefore, band-stop or low-pass filtering through frequency domain analysis can effectively suppress combustion interference, the main noise source, thereby highlighting the low-frequency characteristic signals related to friction. This is crucial for extracting weak fault features from strong background noise.

[0054] The Fast Fourier Transform (FFT) is a mature and efficient time-frequency conversion algorithm. Its function is to convert the time-domain rotational speed fluctuation signal to the frequency domain, obtaining a distribution map of the signal energy across different frequency components, i.e., the frequency spectrum. The frequency spectrum allows for the intuitive identification of frequency bands where energy is concentrated, providing a visual basis for subsequently distinguishing different types of excitation sources. This is a fundamental operation for frequency domain analysis and filtering.

[0055] Identifying the combustion excitation frequency band is crucial for distinguishing between combustion excitation and signals (friction characteristics). For example, for a 4-cylinder, 4-stroke engine operating at 2000 rpm, its ignition frequency is approximately (4*2000) / 120 ≈ 66.7 Hz. Its second harmonic is approximately 133.3 Hz, and its third harmonic is approximately 200 Hz. Therefore, on the frequency spectrum, a relatively wide band (e.g., 50 Hz to 250 Hz) centered at 66.7 Hz, covering its multiple harmonics, can be identified and marked as the "combustion excitation frequency band." This step clarifies the frequency range that needs to be filtered out or avoided.

[0056] After identifying the lower limit of the combustion excitation frequency band (e.g., 50Hz in the example above), all spectral components below this frequency (e.g., 0-45Hz) are extracted. Then, an inverse fast Fourier transform is performed on these retained low-frequency spectral components to convert them back to a time-domain signal. This newly generated time-domain signal has had its high-frequency combustion impact components effectively filtered out, primarily retaining the low-frequency torsional vibration response excited by the imbalance of reciprocating inertial forces and changes in frictional torque, i.e., the "actual friction loss characteristic signal" defined in this scheme. This signal will serve as the "actual fingerprint" for subsequent comparison with model prediction results.

[0057] S130, Input the operating condition parameters of the current operating condition into the pre-established engine tribological system model, and run the model to output the predicted friction torque change pattern when at least one preset friction pair experiences a preset degree of wear under the current operating condition, wherein the engine tribological system model is configured to simulate the friction torque change characteristics of different friction pairs under different degrees of wear.

[0058] The "pre-established engine tribological system model" is obtained through the following offline construction steps: A high-fidelity simulation model is established, incorporating multibody dynamics of the engine crankshaft and connecting rod mechanism and the hydrodynamic lubrication theory of key friction pairs; in the simulation model, an independent and continuously adjustable wear state parameter is defined for each preset friction pair to be monitored. This wear state parameter is used to quantitatively characterize the increase in equivalent lubrication clearance or surface equivalent roughness of the friction pair; under multiple discrete operating conditions covering the typical operating range of the engine, the wear state parameters of each preset friction pair are systematically adjusted, the simulation model is run, and the output total friction torque fluctuation time-series data is recorded; all simulation results are organized to construct a mapping relationship database with combinations of operating condition parameters and wear state parameters as inputs and corresponding friction torque fluctuation characteristic patterns as outputs.

[0059] It is understood that this step initiates a simulation prediction branch based on physical models. Its function is to utilize known engine physical parameters and tribological principles to generate, through computational simulation, the signal characteristics that should theoretically appear under various preset fault assumptions, thus serving as a "theoretical fingerprint" database for diagnosis. Unlike black-box models trained on large amounts of fault data, this is a white-box or gray-box model based on first principles. Its prediction results have clear physical meaning and interpretability, establishing a causal relationship between "a specific wear state of a specific component" and "observable signal characteristics." This is the theoretical foundation for achieving accurate fault location and severity assessment.

[0060] Specifically, model building is done offline, typically requiring the use of specialized commercial simulation software (such as AVLExcite, GT-SUITE) or self-developed programs to establish a flexible multibody dynamics model including components such as the crankshaft, connecting rod, and piston, in order to accurately calculate inertial forces and elastic deformation. Simultaneously, for key friction pairs such as piston ring-cylinder liner, crankshaft main journal-main bearing, and connecting rod big end-connecting rod journal, a Reynolds equation-based hydrodynamic lubrication model or a more advanced hybrid lubrication model is integrated to simulate oil film pressure distribution, friction, and lubrication conditions. This model also needs to integrate engine load boundary conditions (such as in-cylinder combustion pressure curves, usually obtained through thermodynamic cycle simulation or actual measurement).

[0061] To simulate wear failures, physical wear quantities need to be converted into adjustable parameters in the model. An independent wear state parameter is defined for each preset friction pair to be monitored. For example, for the main bearing, this parameter can be defined as the equivalent radial clearance; in the simulation, the increase in clearance due to wear is simulated by proportionally increasing the bearing bore diameter. For the piston ring, this parameter can be defined as the effective radial elasticity attenuation coefficient of the ring (between 0 and 1), or directly as the equivalent roughness between the ring and the cylinder wall; by adjusting this coefficient or roughness value, the changes in sealing and lubrication characteristics caused by ring wear are simulated. For the cam-taper pair, this parameter can be defined as the change in the equivalent radius of curvature of the contact surface or the increment of the friction coefficient.

[0062] Within the engine's entire operating envelope, a series of discrete typical operating points are selected, covering combinations from idle to high speed and from low load to high load. At each fixed operating point, a systematic parameter scan simulation is performed: each preset friction pair is selected sequentially, and its "wear state parameter" is gradually adjusted from its nominal value (representing a brand-new state) to a set upper limit value (representing a moderate or severe wear state) according to a preset step size, while keeping other friction pairs in a nominal healthy state. Each adjustment is followed by a transient dynamic simulation, calculating one complete working cycle or several cycles.

[0063] Each simulation run records high-resolution time-series data of the total engine friction torque fluctuation over time. Subsequently, all simulation results undergo post-processing and standardization. This includes normalizing the time-series data (eliminating excessive dependence of absolute amplitude on operating conditions), feature extraction (which may also preserve the original waveform), and labeling. Finally, a structured mapping database is constructed. Each record in this database is essentially a "{key: value}" pair:

[0064] Input (key): Operating parameters (speed, load) + fault configuration (friction pair ID, wear level).

[0065] Output (numerical): The corresponding, standardized characteristic pattern of frictional torque fluctuation. This pattern can be a feature vector (such as energy in a specific frequency band) or a representative time-series waveform data.

[0066] During online diagnostics, "running the model" does not refer to performing time-consuming real-time simulation calculations, but rather to performing efficient queries and interpolation calculations on a pre-built mapping database.

[0067] The onboard processor takes as input real-time monitored operating parameters (e.g., 1950 rpm, 62% load) and fault assumptions required by diagnostic logic. The system uses the current operating conditions and fault assumptions as query conditions to quickly search the mapping database. Since the operating points are discrete, algorithms such as nearest neighbor matching or bilinear interpolation may be needed to "synthesize" the theoretically predicted waveform under the current precise operating condition from simulation data of the closest known operating points. The output query or interpolation result is the predicted friction torque change pattern that the engine should produce under the current real-world operating conditions, assuming a specific degree of wear on a particular friction pair. This "pattern" serves as the theoretical benchmark for comparison with actual signals.

[0068] S140, perform pattern matching analysis between the actual friction loss characteristic signal and the predicted friction torque change pattern to calculate the pattern matching degree for the preset friction pair. At the same time, based on the carbon monoxide concentration change trend, determine whether there is an abnormal trend in exhaust components that is associated with the decrease in mechanical efficiency.

[0069] The method for calculating the pattern matching degree is to calculate the cross-correlation coefficient, specifically using the following formula:

[0070]

[0071] in, The first characteristic signal of actual friction loss Data from each sampling point, The first of the predicted friction torque variation patterns Data from each sampling point, and These are their means, The total number of sampling points. That is, the cross-correlation number. The numerical range of is [-1, 1], and the closer its absolute value is to 1, the higher the matching degree.

[0072] Specifically, the goal of pattern matching analysis is to quantitatively assess the similarity between the "fingerprint" (actual friction loss characteristic signal) extracted from actual engine signals and the "theoretical fingerprint" (predicted friction torque variation pattern) predicted by the tribological model for a specific fault assumption. Before performing the matching calculation, it is necessary to ensure that the two signal sequences are comparable. This typically includes: length uniformity, phase alignment, and normalization. The normalized cross-correlation coefficient is calculated using the formula described above. This refers to the "pattern matching degree" for the preset friction pair and the preset wear degree. -1 indicates a completely negative correlation (completely opposite shapes); A value of 0 indicates linear independence. The closer the absolute value is to 1, the higher the waveform similarity between the observed friction characteristics and the specific wear pattern predicted by the model.

[0073] The system calculates a corresponding pattern matching degree for each preset friction pair and preset wear degree of interest, thus forming a matching degree spectrum. For example, the matching degree with the "moderate wear of the first main bearing" pattern is 0.85, and the matching degree with the "light wear of the piston ring in the third cylinder" pattern is 0.40. A high matching degree indicates a high probability of the fault hypothesis.

[0074] The determination of whether there is an abnormal trend in exhaust composition associated with a decrease in mechanical efficiency is achieved through the following steps: continuously monitoring the time-series data of carbon monoxide concentration under stable operating conditions; performing linear trend fitting on the time-series data of carbon monoxide concentration in the exhaust composition data to calculate the average upward slope of its concentration over time; comparing the calculated average upward slope with a first threshold obtained by querying a preset dynamic slope threshold table based on the current operating conditions; if the average upward slope is greater than or equal to the first threshold, it is determined that there is an abnormal trend in exhaust composition; wherein, the dynamic slope threshold table stores the verified slope thresholds corresponding to different engine speed and load combinations.

[0075] Only carbon monoxide concentration time series data within the stable diagnostic operating condition window (stable operating condition) are selected to ensure that trend analysis is not affected by drastic fluctuations in transient operating conditions. A linear least-squares fit is performed on the carbon monoxide concentration sequence within the selected time window (e.g., 2 minutes). The slope of the fitted line represents the average rate of change of carbon monoxide concentration during that time period (unit: ppm / second or ppm / minute). A positive slope indicates an upward trend in concentration. The calculated slope is compared with a "first threshold" obtained from a dynamic slope threshold table. This threshold table is established through engine calibration tests: on a large number of healthy engines of the same model, the natural fluctuations of carbon monoxide concentration are monitored over a long period at various steady-state operating points, and the fluctuation range is statistically analyzed (e.g., taking the upper limit of the fluctuation range plus a safety margin), ultimately determining a maximum allowable fluctuation slope threshold for each {speed, load} operating point. Therefore, the "first threshold" is adaptive to the current operating condition. If the slope is greater than or equal to the "first threshold," the "abnormal trend of exhaust components" flag for the main path is determined to be valid.

[0076] Furthermore, the "determination of whether there is an abnormal trend in exhaust composition associated with a decrease in mechanical efficiency" also includes a supplementary verification step: while making a judgment based on the carbon monoxide concentration trend, the real-time air-fuel ratio of the engine is also calculated based on the oxygen concentration data in the exhaust composition data; the calculated real-time air-fuel ratio is compared with the normal theoretical range of air-fuel ratio determined based on the current operating conditions and engine calibration data; if the real-time air-fuel ratio continues to deviate from the normal theoretical range and reaches a preset deviation tolerance, this is used as auxiliary evidence to strengthen the conclusion of the "abnormal trend in exhaust composition".

[0077] Specifically, the engine's real-time air-fuel ratio is calculated using wide-range oxygen sensor signals over the same time period, combined with parameters such as intake air volume (or directly based on the oxygen sensor voltage-air-fuel ratio characteristic curve). This real-time air-fuel ratio is then compared to the normal theoretical range of air-fuel ratio specified by the engine calibration data under steady-state conditions (usually a narrow interval around the target air-fuel ratio). If the real-time air-fuel ratio continuously deviates from the theoretical range (e.g., exceeding a certain time percentage), and the absolute value of the deviation exceeds a preset "deviation tolerance" (e.g., 0.3 air-fuel ratio units lower than the theoretical lower limit), an auxiliary evidence flag is generated. This flag itself does not make a final judgment, but can be used to enhance or weaken the confidence of the main judgment logic conclusion. For example, when the carbon monoxide trend slope is slightly lower than but close to the first threshold, if the air-fuel ratio auxiliary verification flag is also valid, the system may be more inclined to believe that an abnormal trend exists.

[0078] Understandably, the dual-path collaborative judgment mechanism of "mechanical signal matching" and "chemical trend verification" does not simply add mechanical and chemical signals together, but rather allows each signal to independently form evidence through rigorous algorithms, and then performs deep logical-level fusion. This architecture ensures that even if a signal path is briefly disturbed, it will not immediately lead to a misjudgment of the entire system, thereby significantly improving the robustness and reliability of the diagnostic system.

[0079] S150, if and only if the mode matching degree for a certain preset friction pair exceeds the preset matching threshold, and an abnormal trend in exhaust composition is simultaneously determined, an early abnormal wear warning message for the preset friction pair is generated and output, and the warning message includes the identifier of the friction pair being warned.

[0080] The "generating and outputting early abnormal wear warning information for the preset friction pair" also includes a step of quantifying the severity of wear: based on the specific value of the pattern matching degree exceeding the matching threshold, it is mapped to one of the preset severity level ranges; in addition to the friction pair identifier, the warning information also includes a data field to record the determined severity level, thus forming a composite warning information that includes the fault location and the preliminary quantified severity.

[0081] Specifically, the system checks the pattern matching degree of each of the currently monitored preset friction pairs to see if it exceeds a preset matching threshold (e.g., For any candidate friction pair exceeding the threshold, the system simultaneously checks for abnormal trend indicators in the exhaust composition. Only when a friction pair simultaneously meets both of the above conditions is the system considered the early abnormal wear diagnosis for that friction pair valid. This "double insurance" mechanism is the cornerstone of the high reliability of this solution. It can effectively filter out the following two common false alarm scenarios:

[0082] Scenario 1 (False Alarm): Abnormal speed fluctuations caused by non-wear factors such as road impacts or sudden changes in accessory load may temporarily increase the matching degree of a certain friction pair and exceed the threshold. However, since these events do not affect combustion efficiency, the exhaust trend indicator will not be valid, and therefore no warning will be triggered.

[0083] Scenario 2 (Chemical False Alarm): Fluctuations in air-fuel ratio and abnormal CO trends may occur due to fuel batch differences, changes in air quality, or slight drift of the intake sensor, but the engine's mechanical condition remains intact. In this case, although the chemical indicator may be valid, the wear warning will not be triggered because the mechanical matching degree of all friction pairs is within the threshold (or does not simultaneously point to the same specific component).

[0084] To improve the usability of early warning information, the system not only provides qualitative alarms but also performs preliminary quantitative assessments. "Mapping the pattern matching degree to multiple preset severity level ranges based on the specific numerical value exceeding the matching threshold" is a lookup table or conditional judgment process. For example, the following mapping rules can be preset:

[0085] If the matching degree If so, it is mapped to a "low-level" warning.

[0086] If the matching degree If so, it is mapped to a "medium" warning.

[0087] If the matching degree If so, it will be mapped to an "advanced" warning.

[0088] This mapping relationship, calibrated based on simulation and experimental data, reflects the correlation trend between signal similarity and wear degree. The higher the value, the more closely the actual wear pattern matches the severe wear pattern predicted by the model.

[0089] The final generated warning message is a standard, machine-readable structured data object, rather than a simple text prompt. Its typical data structure includes the following fields: friction pair identifier, severity level, timestamp, confidence level, and correlation data.

[0090] The aforementioned "output" means that the structured warning information is published to a designated communication interface or data bus. This can be: displayed on the driver's dashboard or a technician's handheld diagnostic device, providing intuitive text and graphic prompts; and / or sent to the engine control unit via an in-vehicle network (such as the CAN bus) to provide input for possible subsequent adaptive protection controls; and / or stored in a local or remote historical database for long-term health status tracking and trend analysis.

[0091] Understandably, this solution transforms the intelligent diagnostic results from multi-source information fusion into engineering outputs with clear action guidance value. Rigorous dual-verification logic ensures the accuracy of alarms, and through quantitative grading and structured encapsulation, early warning information becomes a key data node connecting fault diagnosis, real-time control, and predictive maintenance decisions, thus achieving a closed loop from theoretical methodology to engineering application.

[0092] Following S150, a closed-loop control step is also included: the generated early abnormal wear warning information is sent to the engine control unit via the vehicle communication network; the engine control unit is pre-programmed to adaptively execute a predefined protective control strategy based on the friction pair identifier and severity level contained in the warning information after receiving it. The protective control strategy includes at least one of the following: specifically limiting the torque output of a specific cylinder, increasing the oil pressure setpoint of the main oil passage or a specific lubrication branch, or activating a pulse-type enhanced lubrication mode for suspected wear areas.

[0093] Specifically, the generated early abnormal wear warning information is broadcast or transmitted point-to-point via standard vehicle network such as the vehicle controller area network (VLAN) in a predefined data frame format (such as CAN protocol data frames). The data frame encapsulates the structured information generated by S150, ensuring its real-time and reliable transmission to the target node—the engine control unit (ECU). The ECU has a pre-set diagnostic message receiving and parsing module. Upon receiving the warning information data frame, the ECU immediately decodes it and extracts key fields, mainly including: identifying the specific affected friction pair and determining the severity level of the abnormality.

[0094] The engine control unit (ECU) has one or more pre-stored fault mode-protection strategy mapping tables or decision trees. Based on the parsed friction pair identifiers and severity levels, the ECU adaptively invokes and executes pre-programmed, optimal protective control strategies. These strategies include:

[0095] Strategy 1: Targeted limitation of torque output in specific cylinders

[0096] Applicable scenarios: Mainly applicable to situations where wear is located on piston ring assemblies, piston pins, or connecting rod big end bearings in specific cylinders.

[0097] Execution mechanism: The ECU reduces the indicated torque of a specific cylinder by adjusting the electronic throttle opening, fuel injection quantity, or ignition advance angle, either individually or in combination.

[0098] Strategy 2: Increase the oil pressure setting value of the main oil passage or specific lubrication branches.

[0099] Applicable scenarios: Suitable for wear warning of bearings (main bearings, connecting rod bearings), or as a general initial protective measure.

[0100] Execution mechanism: For engines equipped with variable displacement oil pumps, the ECU sends a command to the oil pump control system to increase its output pressure setpoint.

[0101] Strategy 3: Activate the pulse-based enhanced lubrication mode targeting suspected wear areas.

[0102] Applicable scenarios: Suitable for engines with additional, independently controllable lubrication injection points (such as piston cooling nozzles, directional bearing injection nozzles).

[0103] Execution mechanism: The ECU controls the solenoid valve or piezoelectric injector to inject an additional, metered amount of engine oil into the area near the friction pair indicated in the warning message during a specific engine cycle phase.

[0104] The above strategies can be adaptively adjusted in intensity or used in combination depending on the severity of the warning:

[0105] Low-level warning: You can simply take "Strategy Two" (slightly increase the oil pressure) and observe.

[0106] Intermediate warning: A combination of "Strategy 1" (mildly limiting the torque of the relevant cylinder) and "Strategy 2" (moderately increasing the oil pressure) can be adopted.

[0107] Advanced warning: It can simultaneously activate "Strategy 1" (stronger torque limiting), "Strategy 2" (higher pressure) and "Strategy 3" (activate enhanced lubrication), and may also provide the driver with clearer and more immediate maintenance reminders.

[0108] Understandably, through the above closed-loop control steps, the system no longer waits for the fault to worsen to the point of severe performance degradation or mechanical damage. Instead, it proactively and precisely adjusts the operating parameters when the diagnostic system detects early signs. While maintaining basic functional output, it prioritizes the mechanical integrity of the engine, truly achieving condition-based predictive maintenance and extending the engine's service life and operational safety.

[0109] In one implementation, the engine anomaly detection method combining a gas sensor and an engineering mechanics model includes a long-term health status prediction step: establishing a historical database to persistently store the warning information generated in each execution step S150, the corresponding pattern matching degree values, and the operating condition parameters at that time; based on the historical database, performing regression analysis on the change sequence of the pattern matching degree values ​​of a specified friction pair over time or the cumulative working mileage of the engine, and calculating its long-term trend slope; if the analysis finds that the long-term trend slope exceeds a preset trend alarm threshold, generating and outputting a predictive maintenance prompt message, which indicates that the performance degradation of the corresponding friction pair is accelerating and suggests prioritizing inspection.

[0110] Specifically, the historical database table includes the following fields: timestamp of data record, cumulative engine running time or mileage, snapshot of operating parameters (usually speed and load) at the time of diagnosis, friction pair identifier, mode matching degree, whether a warning was generated, and the severity level of the warning.

[0111] The system continuously and automatically writes the results of each diagnosis into the database. Data preprocessing can be performed before analysis. For example, records with low confidence levels generated during diagnostic runs under unstable operating conditions (based on pre-screening steps) can be filtered out; and a moving average can be applied to multiple matching degree data of the same friction pair under similar operating conditions to smooth out random fluctuations.

[0112] When a health trend assessment of a specified friction pair (such as the "first main bearing") is required, the system extracts the time series of all records for that friction pair from the historical database. The extracted health indicator series, sorted by time or mileage, is fitted using a linear regression method. The slope of the fitted line represents the average rate of change of the friction pair's health indicator (e.g., the fit increases by an average of 0.05 per 100 hours of operation). A positive and larger slope indicates a faster rate of performance degradation.

[0113] The "trend alarm threshold" is a key engineering parameter, set based on statistical analysis of a large amount of reliability engineering data or historical maintenance data from similar engines. By analyzing the typical degradation process of the friction pairs of this type of engine from slight wear to the need for maintenance, an upper limit for the "normal slow aging" of the matching degree trend slope can be determined. The "trend alarm threshold" is usually set slightly higher than this upper limit to sensitively detect anomalies of "accelerated degradation".

[0114] Initiate trend analysis tasks periodically (e.g., monthly) or as needed. Calculate the long-term trend slope. Will be compared with the preset threshold Compare them.

[0115] like It is believed that the degradation rate of the friction pair is within the expected range and is in the normal aging process.

[0116] like The performance degradation of the friction pair is accelerating, and its remaining service life may be shorter than expected.

[0117] When accelerated degradation is detected, the system automatically generates and outputs a predictive maintenance alert. This alert is highly actionable, and its structure is illustrated below:

[0118] Component: First main bearing.

[0119] Conclusion: The rate of performance degradation exceeds the normal range and shows an accelerating trend.

[0120] Data support: In the past 500 operating hours, the health indicator (matching degree) rose from 0.72 to 0.88, with a trend slope β=0.032 / hundred hours, exceeding the threshold of 0.025 / hundred hours.

[0121] Recommendation: It is recommended that this component be given priority for inspection and condition assessment during the next scheduled maintenance (or within the next XX operating hours).

[0122] Understandably, the value of this long-term health status prediction step lies in its ability to connect discrete, point-like diagnostic alarms into a continuous, linear health evolution trajectory. This transforms maintenance decisions from "response-based" maintenance based on single alarms to "planned" maintenance based on long-term trend predictions. The generated structured alerts can be directly integrated into fleet management software or maintenance work order systems to guide optimal allocation of maintenance resources, maximizing component lifespan while ensuring safety, and achieving an optimal balance between maintenance costs and reliability. This marks the evolution of the entire system from a "precise diagnostic instrument" to a "full lifecycle health management partner."

[0123] This invention, by employing the aforementioned technical solution, offers the following beneficial effects: By collaboratively verifying the wear characteristics of specific friction pairs analyzed from speed fluctuations with the systematic mechanical efficiency decline trend extracted from exhaust components, a dual-insurance diagnostic logic is established. This not only effectively filters out false alarms caused by single-signal interference, significantly improving diagnostic reliability, but more importantly, by matching actual signals with a theoretical fault fingerprint database generated from a high-fidelity tribological model, it achieves a leap from 'whether there is a fault' to 'which component and to what extent the fault exists,' providing precise information for predictive maintenance that can be acted upon directly. Simultaneously, the introduction of closed-loop control and long-term trend prediction transforms this solution from a detection tool into a complete active engine health management system.

[0124] The following will illustrate a method for detecting engine anomalies using a gas-sensitive sensor combined with an engineering mechanics model, based on specific embodiments of the present invention.

[0125] This embodiment provides an engine anomaly detection method combining a gas-sensitive sensor and an engineering mechanics model, applied to an online health monitoring system for an inline four-cylinder turbocharged gasoline engine. The system hardware includes: a high-precision magnetoelectric crankshaft position sensor (for acquiring instantaneous speed signals), a wide-range oxygen sensor mounted on the exhaust manifold (for obtaining oxygen concentration to calculate the air-fuel ratio), a non-dispersive infrared carbon monoxide sensor, and an onboard embedded processor (as a diagnostic calculation unit).

[0126] The implementation steps of the method are as follows:

[0127] The onboard processor continuously monitors engine speed and load signals (e.g., calculated based on throttle opening and intake air volume). It calculates the fluctuations in speed and load within the most recent fixed time window (e.g., 5 seconds). The engine is considered to have entered a "stable diagnostic window" only when the speed fluctuation does not exceed ±30 rpm and the load fluctuation does not exceed ±5%. Within this window, the onboard processor simultaneously acquires a high-precision sequence of instantaneous crankshaft speed fluctuations (obtained by measuring the flywheel tooth passage time) and exhaust carbon monoxide concentration values. If the engine has not entered a stable window, only the data is cached, and the diagnostic process is not performed.

[0128] Furthermore, the onboard processor processes a segment of instantaneous speed fluctuation signal acquired within a stable window. First, a Fast Fourier Transform (FFT) is performed on the signal to obtain its spectrum. For this four-cylinder engine, at 2000 rpm, its ignition frequency (combustion excitation frequency) is approximately 66.7 Hz. The frequency band centered at 66.7 Hz, containing its second and third harmonics (approximately 133.3 Hz and 200 Hz), is identified on the spectrum and marked as the "combustion excitation frequency band." Subsequently, all spectral components with frequencies below the lower limit of this frequency band (e.g., 50 Hz) are extracted. Finally, an Inverse Fast Fourier Transform (IFFT) is performed on these low-frequency components to reconstruct a time-domain signal. This signal filters out direct combustion impact noise and mainly contains low-frequency torsional vibration components excited by the imbalance of reciprocating inertial forces and the fluctuation of friction torque of each friction pair, which is the "actual friction loss characteristic signal" defined in this scheme.

[0129] It should be noted that the model prediction and theoretical fingerprint retrieval rely on an offline, pre-built "engine tribological system model." This model is developed based on commercial multibody dynamics and fluid lubrication co-simulation software. The model refines the dynamics and lubrication interfaces of the crankshaft-main bearing system and the piston assembly-cylinder liner system. "Wear state parameters," such as the piston ring equivalent tension coefficient and the main bearing equivalent radius clearance, are defined for the four preset friction pairs to be monitored (in this example, the piston ring assembly of the first cylinder, the first main bearing, the second main bearing, and the third main bearing).

[0130] During the model development phase, engineers systematically adjusted the wear state parameters of each friction pair at hundreds of speed-load conditions covering the engine's operating range (simulating different states from brand new to moderate wear), ran simulations, and recorded the time-series curve of the engine's total friction torque fluctuation for each simulation. Ultimately, all simulation results were compiled into a multidimensional database (mapping relation library), with inputs of "operating condition parameters + friction pair identifier + wear degree" and outputs of the corresponding "predicted friction torque change pattern".

[0131] During online diagnostics, the onboard processor inputs the current stable operating parameters (e.g., 2000 rpm, 65% load) into the model system. The model system does not perform real-time simulation; instead, it acts as a query interface. Based on the current operating conditions and preset comparison requirements (e.g., querying the theoretical signals when the four friction pairs are at the "slight wear" level), it quickly retrieves and outputs four corresponding "predicted friction torque change pattern" waveforms from the mapping database as theoretical comparison benchmarks.

[0132] When performing dual-path information fusion and comprehensive judgment, two analysis threads are executed in parallel.

[0133] Thread 1 (Mechanical Side Pattern Matching): The onboard processor calculates the normalized cross-correlation coefficient between the "actual friction loss characteristic signal" and each "predicted friction torque change pattern," using the following formula:

[0134]

[0135] in, The first characteristic signal of actual friction loss Data from each sampling point, The first of the predicted friction torque variation patterns Data from each sampling point, and These are their means, The total number of sampling points. That is, the cross-correlation number. The numerical range of is [-1, 1], and the closer its absolute value is to 1, the higher the matching degree.

[0136] The calculation results are as follows: the correlation coefficient with the "slight wear of the piston ring assembly in the first cylinder" mode is 0.84, the correlation coefficient with the "slight wear of the first main bearing" mode is 0.45, and the correlation coefficients with the other modes are all below 0.3. The preset matching threshold is 0.70. Therefore, the mode matching degree of "piston ring assembly in the first cylinder" (0.84) significantly exceeds the threshold.

[0137] Thread 2 (Chemical Trend Verification): The onboard processor analyzes the change in carbon monoxide concentration over time within the same stable diagnostic window. A linear fit is performed on the carbon monoxide concentration time series to obtain its average upward slope. The onboard processor queries a preset "dynamic slope threshold table" based on the current operating conditions (2000 rpm, 65% load). This threshold table is calibrated using test data from a large number of healthy engines under the same operating conditions, and the first threshold corresponding to this condition is... It is 2.0 ppm / min. If the calculated... If the result is negative, it is determined that there is an "abnormal trend in exhaust composition". As a supplementary verification, the system also calculates the real-time air-fuel ratio obtained from the wide-range oxygen sensor and finds that it is consistently slightly lower than the theoretical optimal air-fuel ratio range under this operating condition, further confirming the slight deterioration in combustion efficiency.

[0138] The onboard processor executes the final decision logic: the diagnostic conclusion is confirmed if and only if both the pattern matching degree of the "first cylinder piston ring assembly" (0.84 > 0.70) and the "abnormal trend of exhaust composition" flag (True) are simultaneously true. Based on the preset range where the matching degree of 0.84 falls (e.g., 0.8-0.9 corresponds to "intermediate"), the system generates a structured warning message:

[0139] {Faulty component: Piston ring assembly of cylinder 1; Severity: Medium; Timestamp: [Current Time]; Confidence: High}.

[0140] The warning message is sent to the engine control unit via the CAN bus. The pre-programmed control strategy in the ECU is triggered. In response to the warning about the "piston ring assembly of cylinder 1", the ECU performs the following in one implementation: slightly reduces the upper limit of the fuel injection quantity of cylinder 1 (to achieve targeted torque limitation) and increases the oil pump pressure by one level, in order to delay the wear process by improving lubrication.

[0141] All early warning information and corresponding matching degree and operating condition data are stored in the historical database. The system periodically performs trend analysis on the historical matching degree data of the "first cylinder piston ring assembly". If it is found that the matching degree has increased linearly from 0.75 to 0.88 in the past 100 operating hours, and the trend slope exceeds the preset "accelerated decay" alarm threshold, the system will generate a predictive maintenance prompt: "The wear of the first cylinder piston ring assembly is showing an accelerated trend, and priority inspection is recommended", thus realizing the sublimation from fault warning to life prediction.

[0142] Through the above technical solution, this invention achieves intelligent, high-confidence diagnosis and early warning of the early wear state of key friction pairs in engines. This solution constructs a closed-loop diagnostic system by integrating two-dimensional information from mechanical dynamics and combustion chemistry. The system first accurately separates the characteristic components representing the friction state from the complex raw speed signal, and simultaneously generates theoretical fingerprints corresponding to different fault modes using a high-fidelity tribological model. Then, it matches the real-time extracted features with the theoretical fingerprints and performs collaborative verification by combining this with exhaust component trends reflecting overall mechanical efficiency. This dual-path information fusion mechanism enables the system to effectively suppress interference from a single signal source, thereby accurately identifying specific worn components at an early stage. Ultimately, the diagnostic results not only generate quantitative early warning information to guide maintenance but also trigger the control system to implement proactive protection, achieving a complete technical closed loop from state perception and intelligent diagnosis to proactive health management.

[0143] The detailed description of the above specific embodiments fully illustrates the feasibility, preferred implementation, and technical effects achieved by the technical solution of the present invention. Those skilled in the art can make several modifications and substitutions based on the above description without departing from the principles and spirit of the present invention, and these modifications and substitutions should also be considered within the scope of protection of the present invention.

[0144] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention. The actual method is not limited to this. In conclusion, if those skilled in the art are inspired by this description and design similar methods and embodiments without departing from the spirit of the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A method for detecting engine anomalies using a gas-sensitive sensor combined with an engineering mechanics model, characterized in that, The method is executed by an onboard processor and includes: S110, acquire the instantaneous speed fluctuation signal and exhaust composition data of the engine under the current operating conditions, wherein the instantaneous speed fluctuation signal is collected by a high-precision crankshaft speed sensor and reflects the instantaneous angular velocity change of the flywheel gear, and the exhaust composition data includes at least the carbon monoxide concentration collected by a gas-sensitive sensor; S120, perform frequency domain analysis on the instantaneous speed fluctuation signal to separate the low-frequency component that does not overlap with the combustion excitation frequency band, thereby obtaining the actual friction loss characteristic signal that reflects the changes in reciprocating inertial force and friction torque; S130, Input the operating condition parameters of the current operating condition into the pre-established engine tribological system model, and run the model to output the predicted friction torque change pattern when at least one preset friction pair experiences a preset degree of wear under the current operating condition, wherein the engine tribological system model is configured to simulate the friction torque change characteristics of different friction pairs under different degrees of wear. S140, perform pattern matching analysis on the actual friction loss characteristic signal and the predicted friction torque change pattern to calculate the pattern matching degree for the preset friction pair. At the same time, based on the carbon monoxide concentration change trend, determine whether there is an abnormal trend of exhaust components associated with the decrease in mechanical efficiency. S150, if and only if the mode matching degree for a certain preset friction pair exceeds a preset matching threshold, and an abnormal trend of the exhaust composition is simultaneously determined, an early abnormal wear warning message for the preset friction pair is generated and output, the warning message containing the identifier of the friction pair being warned.

2. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, The S120 mid-frequency domain analysis specifically includes: The frequency spectrum is obtained by performing a fast Fourier transform on the instantaneous speed fluctuation signal; Identify the combustion excitation frequency bands of the engine ignition frequency and its harmonics in the frequency spectrum; Extract the spectral components in the frequency spectrum that are lower than the lowest frequency of the combustion excitation band and perform an inverse transformation to obtain the actual friction loss characteristic signal that characterizes the changes in reciprocating inertial force and friction torque.

3. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, In S140, the method for calculating the pattern matching degree is to calculate the cross-correlation coefficient, specifically using the following formula: in, The first characteristic signal of actual friction loss Data from each sampling point, The first of the predicted friction torque variation patterns Data from each sampling point, and These are their means, The total number of sampling points. That is, the cross-correlation number. The numerical range of is [-1, 1], and the closer its absolute value is to 1, the higher the matching degree.

4. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, The "determination of whether there is an abnormal trend in exhaust composition associated with a decrease in mechanical efficiency" described in S140 is achieved through the following steps: Continuously monitor the time-varying sequence data of carbon monoxide concentration under stable operating conditions; Linear trend fitting was performed on the time sequence data of carbon monoxide concentration to calculate the average slope of its concentration increase over time. The calculated average rising slope is compared with a first threshold obtained by querying a preset dynamic slope threshold table based on the current operating conditions. If the average upward slope is greater than or equal to the first threshold, it is determined that there is an abnormal trend in the exhaust composition. The dynamic slope threshold table stores verified slope thresholds corresponding to different engine speed and load combinations.

5. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, The "pre-established engine tribological system model" was obtained through the following offline construction steps: Establish a high-fidelity simulation model that incorporates the multibody dynamics of the engine crankshaft and connecting rod mechanism and the hydrodynamic lubrication theory of key friction pairs; In the simulation model, an independent and continuously adjustable wear state parameter is defined for each preset friction pair to be monitored. The wear state parameter is used to quantitatively characterize the increase in the equivalent lubrication clearance or the increase in the equivalent surface roughness of the friction pair. Under multiple discrete operating conditions covering the typical operating range of the engine, the wear state parameters of each preset friction pair are systematically adjusted, the simulation model is run, and the output total friction torque fluctuation time series data is recorded. All simulation results were compiled to construct a mapping database with the combination of operating parameters and wear state parameters as input and the corresponding friction torque fluctuation characteristic mode as output.

6. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, The "generating and outputting early abnormal wear warning information for the preset friction pair" described in S150 also includes a step of quantifying the severity of wear: Based on the specific numerical value of the pattern matching degree exceeding the matching threshold, it is mapped to one of a preset range of severity levels; In addition to the friction pair identifier, the warning information also includes a data field for recording the determined severity level, thus forming a composite warning information that includes the fault location and a preliminary quantification of severity.

7. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, Before executing S110, a pre-screening step for operating conditions is also included: The engine speed and load signals are monitored in real time, and their fluctuations within a preset time window are calculated. Only when the speed fluctuation falls within the first preset range and the load fluctuation falls within the second preset range is it determined that the engine has entered the stable diagnostic condition window that can be used for accurate diagnosis, and S110 to S150 are triggered. If the stable diagnostic condition window is not entered, the diagnostic process will be paused or only data caching and recording will be performed.

8. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 4, characterized in that, The step of "determining whether there is an abnormal trend in exhaust composition associated with a decrease in mechanical efficiency" also includes a supplementary verification step: While making a judgment based on the carbon monoxide concentration trend, the engine's real-time air-fuel ratio is also calculated based on the oxygen concentration data in the exhaust component data. The calculated real-time air-fuel ratio is compared with the normal theoretical range of air-fuel ratio determined based on the current operating conditions and engine calibration data; If the real-time air-fuel ratio continues to deviate from the normal theoretical range and reaches a preset deviation tolerance, this will be used as supplementary evidence to strengthen the conclusion of the "abnormal trend of exhaust composition".

9. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, S150 is followed by a closed-loop control step: The generated early abnormal wear warning information is sent to the engine control unit via the vehicle communication network; The engine control unit is pre-programmed to adaptively execute a predefined protective control strategy based on the friction pair identifier and severity level contained in the warning information after receiving the warning information. The protective control strategy includes at least one of the following: specifically limiting the torque output of a specific cylinder, increasing the oil pressure setpoint of the main oil passage or a specific lubrication branch, or activating a pulse-type enhanced lubrication mode for suspected wear areas.

10. The engine anomaly detection method based on a gas-sensitive sensor and an engineering mechanics model according to claim 1, characterized in that, The method also includes a long-term health status prediction step: Establish a historical database to persistently store the warning information, corresponding pattern matching degree values, and operating condition parameters generated in each execution step S150. Based on the historical database, regression analysis is performed on the change sequence of the pattern matching degree value of the specified friction pair with time or the cumulative working mileage of the engine, and the slope of its long-term trend is calculated. If the analysis finds that the slope of the long-term trend exceeds a preset trend alarm threshold, a predictive maintenance prompt message is generated and output. This message indicates that the performance degradation of the corresponding friction pair is accelerating and suggests prioritizing its inspection.

Images