CN117072293A - Light automobile exhaust fault diagnosis method and system - Google Patents

Abstract

The invention provides a light automobile exhaust fault diagnosis method and system, and belongs to the technical field of fault diagnosis. Firstly, acquiring various state parameters in an exhaust system; these state parameters are then converted into standard signal parameters by a signal processing operation. Combining the standard signal parameters again to form fault diagnosis data including data of various exhaust system components; and then the fault diagnosis data are input into a fault intelligent diagnosis model, and the model is analyzed and optimized to generate a final fault intelligent diagnosis model. When real-time data is input into this final model, diagnostic analysis is performed, and a failure evaluation result is output. The invention constructs a novel intelligent diagnosis system which is suitable for research and development and can provide convenient service for consumers, is the future of the development of the after-market of the automobile, develops more high-quality service for the after-market of the automobile, and simultaneously upgrades the protection navigation for the technology of the automobile industry.

Description

Light automobile exhaust fault diagnosis method and system

Technical Field

The invention belongs to the technical field of fault diagnosis, and particularly relates to a light automobile exhaust fault diagnosis method and system.

Background

In order to solve the problem that the development of the automobile industry causes huge pressure on environmental protection, the national standard of six A is implemented in 7 months of 2020, and the stricter standard of six B is implemented in 7 months of 2023, so that the failure of automobile exhaust system caused by failure of six automobile exhaust systems is one of the problems frequently encountered in vehicle maintenance. The failure causes are more, including direct failures such as aging and blocking of a particulate matter trap (GPF), failure of a differential pressure sensor, degradation of an oxygen sensor, failure and congestion of a three-way catalyst, and abnormal emission caused by indirect reasons of an engine electric control system. In traditional automobile maintenance, maintenance operators assist detecting instrument to detect vehicle faults according to experience, the process is tedious, the detection time is long, the labor cost is high, the detection efficiency and the accuracy rate cannot be guaranteed, and especially the failure cause of an exhaust system after technology upgrading is more complex, for example, the phenomenon that an indicator light is frequently turned on still occurs after GPF is subjected to active and passive regeneration, an automobile owner can only spend more time and cost on automobile maintenance, the use experience of consumers to China six automobiles is greatly reduced, and the difficulty of vehicle maintenance mechanism to vehicle fault maintenance is also increased.

Therefore, the design of the intelligent diagnosis method for the exhaust faults of the automobiles is of great significance for development of new standard automobiles in China and maintenance efficiency and quality of six automobiles in China.

Disclosure of Invention

Based on the technical problems, the invention provides a light automobile exhaust fault diagnosis method and system, which acquire emission state parameters according to a nondestructive testing principle, consider nonlinearity and multisource of fault factors, use deep learning to construct a fault intelligent diagnosis model, and provide an effective, quick, convenient and safe scheme for automobile after-market enterprises in diagnosing automobile exhaust system faults.

The invention provides a light automobile exhaust fault diagnosis method, which comprises the following steps:

step S1: acquiring state parameters;

step S2: performing signal processing operation on the state parameters to obtain standard signal parameters;

step S3: combining the standard signal parameters to obtain fault diagnosis data;

step S4: inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model;

step S5: and inputting the real-time data into a final fault intelligent diagnosis model for diagnosis to obtain a fault evaluation result.

Optionally, the acquiring the status parameter specifically includes:

the state parameters include a GPF differential pressure signal, a GPF temperature signal, a GPF load signal, a differential pressure sensor signal, an engine speed signal, a throttle opening signal, an air-fuel ratio sensor signal, a post-oxygen sensor signal, a fuel correction signal, a TWC inlet temperature signal, a TWC outlet temperature signal, a TWC conversion efficiency signal, a post exhaust flow rate signal, a load signal, an ignition coil signal, a spark plug signal, a crankshaft position signal, an exhaust gas pressure signal, and an exhaust pipe vibration acceleration signal.

Optionally, the step of combining the standard signal parameters to obtain fault diagnosis data specifically includes:

the fault diagnosis data comprises GPF fault data, differential pressure sensor fault data, oxygen sensor data, TWC aging data, post exhaust flow rate data, engine misfire data and exhaust pipe vibration data;

the GPF fault data comprises a standard GPF differential pressure signal, a standard GPF temperature signal and a standard GPF load signal; the differential pressure sensor fault data comprises a standard differential pressure sensor signal, a standard engine speed signal and a standard throttle opening signal; the oxygen sensor data includes a standard air-fuel ratio sensor signal, a standard post-oxygen sensor signal, a standard fuel correction signal, and a standard engine speed signal; the TWC aging data includes a standard TWC inlet temperature signal, a standard TWC outlet temperature signal, and a standard TWC conversion efficiency signal; the rear exhaust flow rate data includes a standard rear exhaust flow rate signal, a standard load signal, and a standard engine speed signal; the engine misfire data includes a standard ignition coil signal, a standard spark plug signal, a standard crankshaft position signal, and a standard engine speed signal; the exhaust pipe vibration data comprises a standard tail gas pressure signal, a standard exhaust pipe vibration acceleration signal and a standard engine speed signal.

Optionally, the inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model specifically includes:

the fault intelligent diagnosis model comprises a regression model and a classification model;

respectively inputting oxygen sensor fault data, TWC aging data and rear exhaust gas flow rate data into the regression model for analysis to obtain a final regression model;

and respectively inputting GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data into the classification model for analysis to obtain a final classification model.

Optionally, the oxygen sensor fault data, TWC aging data and post-exhaust flow rate data are respectively input into the regression model for analysis, so as to obtain a final regression model, which specifically includes:

the regression model comprises a first full-connection layer, a second activation function layer, a third full-connection layer, a third activation function layer, a fourth full-connection layer, a fourth activation function layer, a fifth full-connection layer, a fifth activation function layer, a sixth full-connection layer, a sixth activation function layer, a first attention layer, an element-by-element addition layer, a seventh full-connection layer and a seventh activation function layer;

And respectively and sequentially inputting oxygen sensor fault data, TWC aging data and post-exhaust flow velocity data into the first full-connection layer, the first activation function layer, the second full-connection layer, the second activation function layer, the third full-connection layer, the third activation function layer, the fourth full-connection layer, the fourth activation function layer, the fifth full-connection layer, the fifth activation function layer, the sixth full-connection layer, the sixth activation function layer, the attention layer I, the element-by-element addition layer, the seventh full-connection layer and the seventh activation function layer to analyze, thereby obtaining a final regression model.

Optionally, the step of inputting the GPF fault data, the differential pressure sensor fault data, the engine misfire data and the exhaust pipe vibration data into the classification model for analysis respectively to obtain a final classification model specifically includes:

the classification model comprises an eighth full-connection layer, an eighth activation function layer, a ninth full-connection layer, a ninth activation function layer, a tenth full-connection layer, a tenth activation function layer, an eleventh full-connection layer, an eleventh activation function layer, a first splicing layer, a second attention layer, a second splicing layer, a twelfth full-connection layer, a twelfth activation function layer, a thirteenth full-connection layer and a thirteenth activation function layer;

GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data are sequentially input into the eighth full-connection layer, the eighth activation function layer, the ninth full-connection layer, the ninth activation function layer, the tenth full-connection layer, the tenth activation function layer, the eleventh full-connection layer, the eleventh activation function layer, the first splicing layer, the second attention layer, the second splicing layer, the twelfth full-connection layer, the twelfth activation function layer, the thirteenth full-connection layer and the thirteenth activation function layer respectively, and analysis is carried out to obtain a final classification model.

The present invention also provides a light automobile exhaust fault diagnosis system, the system comprising:

the state parameter acquisition module is used for acquiring state parameters;

the signal processing module is used for performing signal processing operation on the state parameters to obtain standard signal parameters;

the signal parameter combination module is used for combining the standard signal parameters to obtain fault diagnosis data;

the fault intelligent diagnosis model identification module is used for inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model;

And the test evaluation module is used for inputting the real-time data into the final fault intelligent diagnosis model for diagnosis to obtain a fault evaluation result.

Optionally, the fault intelligent diagnosis model identification module specifically includes:

the regression model identification submodule is used for respectively inputting the oxygen sensor fault data, the TWC aging data and the rear exhaust flow velocity data into a regression model for analysis to obtain a final regression model;

and the classification model identification sub-module is used for respectively inputting the GPF fault data, the differential pressure sensor fault data, the engine fire data and the exhaust pipe vibration data into the classification model for analysis to obtain a final classification model.

Optionally, the regression model identification submodule specifically includes:

the regression model building unit is used for inputting the oxygen sensor fault data, the TWC aging data and the rear exhaust flow velocity data into the first full-connection layer, the first activation function layer, the second full-connection layer, the second activation function layer, the third full-connection layer, the third activation function layer, the fourth full-connection layer, the fourth activation function layer, the fifth full-connection layer, the fifth activation function layer, the sixth full-connection layer, the sixth activation function layer, the attention layer one, the element-by-element addition layer, the seventh full-connection layer and the seventh activation function layer in sequence respectively for analysis, so that a final regression model is obtained.

Optionally, the classification model identification sub-module specifically includes:

the classification model building unit is used for inputting GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data into the eighth full-connection layer, the eighth activation function layer, the ninth full-connection layer, the ninth activation function layer, the tenth full-connection layer, the tenth activation function layer, the eleventh full-connection layer, the eleventh activation function layer, the first splicing layer, the second attention layer, the second splicing layer, the twelfth full-connection layer, the twelfth activation function layer, the thirteenth full-connection layer and the thirteenth activation function layer in sequence respectively for analysis, and obtaining a final classification model.

Compared with the prior art, the invention has the following beneficial effects:

according to the nondestructive testing principle, various state parameters are collected and subjected to signal processing, and the method can comprehensively and accurately diagnose the faults of the exhaust system and help quickly find out the root cause of the problem; an intelligent diagnosis model is adopted, and the method is respectively suitable for regression and classification analysis. This allows optimal diagnostic results to be obtained for different types of faults; by outputting the fault evaluation result, the vehicle owner and the maintenance personnel can know the health condition of the exhaust system, and are helpful for making a maintenance plan and a repair strategy; and the fault of the exhaust system is diagnosed and solved early, so that the emission of the vehicle can be kept up to the standard, the fuel efficiency is improved, and good drivability is maintained.

Drawings

FIG. 1 is a flow chart of a method for diagnosing exhaust failure of a light automobile according to the present invention;

FIG. 2 is a regression model network architecture diagram of a fault intelligent diagnosis model in a light automobile exhaust fault diagnosis method of the invention;

FIG. 3 is a network structure diagram of a classification model of a fault intelligent diagnosis model in a light automobile exhaust fault diagnosis method of the invention;

fig. 4 is a block diagram of an exhaust failure diagnosis system for a light vehicle according to the present invention.

Detailed Description

The invention is further described below in connection with specific embodiments and the accompanying drawings, but the invention is not limited to these embodiments.

Example 1

As shown in fig. 1, the present invention discloses a light automobile exhaust fault diagnosis method, which comprises the following steps:

step S1: and acquiring a state parameter.

Step S2: and performing signal processing operation on the state parameters to obtain standard signal parameters.

Step S3: and combining the standard signal parameters to obtain fault diagnosis data.

Step S4: and inputting the fault diagnosis data into the fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model.

Step S5: and inputting the real-time data into a final fault intelligent diagnosis model for diagnosis to obtain a fault evaluation result.

The steps are discussed in detail below:

step S1: the method for acquiring the state parameters specifically comprises the following steps:

the state parameters include a vehicle particulate filter (GPF) pressure differential signal, a GPF temperature signal, a GPF load signal, a pressure differential sensor signal, an engine speed signal, a throttle opening signal, an air-fuel ratio sensor signal, a post-oxygen sensor signal, a fuel correction signal, a three-way catalytic converter (TWC) inlet temperature signal, a TWC outlet temperature signal, a TWC conversion efficiency signal, a post exhaust flow rate signal, a load signal, an ignition coil signal, a spark plug signal, a crankshaft position signal, an exhaust pressure signal, and an exhaust pipe vibration acceleration signal.

In this embodiment, the state parameters are typically measured by various sensors. A sensor is a device capable of converting a physical quantity into an electrical or other measurable signal for collecting real-time data to monitor the status of various systems of a vehicle. In an automobile exhaust fault diagnostic method, these sensors are used to collect various parameters in order to monitor the operation of the exhaust system in real time and to help detect potential faults.

In the present embodiment, a differential pressure sensor is used to measure the differential pressure across the particulate filter, an engine speed sensor is used to measure the speed of the engine, and an air-fuel ratio sensor is used to measure the oxygen content in the exhaust gas, etc. Each sensor is specifically designed to measure a specific physical quantity and then convert it into an electrical or other type of signal for analysis and processing by the system.

GPF differential pressure signal, which is a device for capturing particulate matter in engine emissions, is used to monitor whether a blockage occurs inside a GPF, which measures the differential pressure across a particulate matter filter (GPF).

The GPF temperature signal measures the temperature of the particulate filter, which parameter helps to assess the operating state of the GPF, e.g. whether an appropriate operating temperature is reached for combustion of the particulate matter.

The GPF load signal is indicative of an engine load condition that affects the generation of particulate matter and the operation of the GPF.

The differential pressure sensor signal is a sensor signal for monitoring the pressure difference in the exhaust pipe and is used for judging the circulation condition of an exhaust system.

Engine speed signal, measuring engine speed, is the basis of many fault determinations, as different faults can have an impact on speed.

The throttle opening signal, which measures the opening degree of the throttle, is correlated with the engine load and the combustion state, and can be used to evaluate the operating state of the engine.

The air-fuel ratio sensor signals are used for measuring the ratio of air to fuel in exhaust gas and are used for judging whether the combustion state is normal or not and detecting the rich combustion and lean combustion conditions in the combustion process.

The post oxygen sensor signal is used for monitoring the oxygen content in the exhaust gas and judging the combustion efficiency and the working state of the catalytic converter, and can detect the oxygen content in the combustion process so as to judge whether the expected combustion effect is achieved.

The fuel correction signal, which adjusts the signal for the engine fuel supply, can help to detect if the fuel supply is appropriate and if there is a fuel system problem.

The TWC inlet temperature signal, the inlet temperature of a three way catalytic converter (TWC) is measured, and the TWC is a means for reducing emissions of harmful substances in exhaust gas.

The TWC outlet temperature signal measures an outlet temperature of the TWC, the outlet temperature being related to an operating condition and efficiency of the TWC.

The TWC conversion efficiency signal, which refers to the ability of the TWC to convert harmful gases to harmless substances, is estimated from the difference between the inlet and outlet temperatures.

And (5) a rear exhaust flow rate signal, measuring the flow rate of exhaust, and judging the circulation condition and the smoothness of an exhaust system.

The load signal indicates a vehicle load condition, and relates to an engine operating state and a load generated by exhaust gas.

Ignition coil signals are associated with engine ignition systems that affect the combustion process and exhaust gas quality.

The spark plug signal, which is related to the ignition system and combustion conditions, is an important component of the ignition system.

A crankshaft position signal measures a position of an engine crankshaft for synchronizing other parameters and determining an operating state of the engine.

The exhaust pressure signal, which measures the pressure in the exhaust pipe, is related to exhaust flow and back pressure.

And the vibration acceleration signal of the exhaust pipe detects the vibration condition of the exhaust pipe and is related to the structural problem of an exhaust system.

Step S2: the signal processing operation is carried out on the state parameters to obtain standard signal parameters, which concretely comprise:

the raw state parameters acquired from the sensor are affected by noise, and filtering technology such as mean filtering, median filtering or Gaussian filtering is used to smooth the signal and remove the noise, so that the signal quality is higher; useful features are extracted from the processed signal. For example, extracting the maximum speed, the average speed from the engine speed signal, and the rate of change from the temperature signal, these features can better represent important information of the signal; for signal parameters in different ranges and units, normalization or standardization operation is carried out, so that all the characteristics are ensured to be on the same scale, the convergence and training effects of the neural network model are facilitated, if the number of state parameters is large, the high-dimensional characteristics are converted into lower dimensions by using Principal Component Analysis (PCA), so that the computational complexity is reduced, and redundant information is eliminated; for a class type parameter, such as a sensor identifier or status indication, an encoding operation, such as a single-hot encoding, is required so that the neural network can understand these non-numeric features; for signals needing to analyze frequency information, such as vibration signals, the signals can be converted from a time domain to a frequency domain through Fourier transformation so as to obtain frequency spectrum information; and processing missing data, filling missing values by using an interpolation method, and ensuring the integrity of input features.

Step S3: combining the standard signal parameters to obtain fault diagnosis data, which specifically comprises the following steps:

in this embodiment, the fault diagnostic data includes GPF fault data, differential pressure sensor fault data, oxygen sensor data, TWC aging data, post exhaust flow rate data, engine misfire data, and exhaust pipe vibration data.

In this embodiment, the GPF fault data includes a standard GPF differential pressure signal, a standard GPF temperature signal, and a standard GPF load signal; the differential pressure sensor fault data comprises a standard differential pressure sensor signal, a standard engine speed signal and a standard throttle opening signal; the oxygen sensor data includes a standard air-fuel ratio sensor signal, a standard post-oxygen sensor signal, a standard fuel correction signal, and a standard engine speed signal; the TWC aging data includes a standard TWC inlet temperature signal, a standard TWC outlet temperature signal, and a standard TWC conversion efficiency signal; the post exhaust flow rate data includes a standard post exhaust flow rate signal, a standard load signal, and a standard engine speed signal; the engine misfire data includes a standard ignition coil signal, a standard spark plug signal, a standard crankshaft position signal, and a standard engine speed signal; the exhaust pipe vibration data includes a standard exhaust gas pressure signal, a standard exhaust pipe vibration acceleration signal, and a standard engine speed signal.

In FIGS. 2-3, dense [ ]) Representing the total connection layer is->Individual nervesA meta-element; activation function layer ([ (Relu), (Linear), (Softmax)])；FR/>Representing the tensors obtained in the regression model, < >>The value range is [1,16 ]]，/>Is an integer; FC->Representing the tensors obtained in the regression model, < >>The value range is [1,15 ]]，/>Is an integer; add (/ -)>，/>) Representation->，Performing element-by-element addition; concat (+)>，/>) Representation->，/>And (5) splicing.

Step S4: inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model, which specifically comprises the following steps:

step S41: and respectively inputting the oxygen sensor fault data, the TWC aging data and the rear exhaust gas flow rate data into a regression model for analysis to obtain a final regression model.

As shown in fig. 2, step S41 specifically includes:

sequentially inputting oxygen sensor fault data, TWC aging data and rear exhaust flow speed data into a first full-connection layer and a first activation function layer respectively to perform full-connection and activation operations to obtain a one-dimensional tensor FR2 (16,); the one-dimensional tensor FR2 contains 16 elements; inputting the one-dimensional tensor FR2 into a second full-connection layer and a second activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FR4 (32,); the one-dimensional tensor FR4 contains 32 elements; inputting the one-dimensional tensor FR4 into a third full-connection layer and a third activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FR6 (64,); the one-dimensional tensor FR6 contains 64 elements; inputting the one-dimensional tensor FR6 into a fourth full-connection layer and a fourth activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FR8 (128); the one-dimensional tensor FR8 contains 128 elements; inputting the one-dimensional tensor FR8 into a fifth full-connection layer and a fifth activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FR10 (256); the one-dimensional tensor FR10 contains 256 elements; inputting the one-dimensional tensor FR10 into a sixth full-connection layer and a sixth activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FR12 (512,); the one-dimensional tensor FR12 contains 512 elements; input the one-dimensional tensor FR12 to the attention layer one through the weight matrix ，/>And->FR12 is subjected to linear transformation to obtain a linear transformation result +.>，/>And->Calculate the query->And bond->The similarity score between is achieved by calculating a dot product (inner product), which can reflect the degree of association between the query and the key, with a higher score indicating a more relevant relationship between the two, and normalizing the score to an attention weight using a Softmax function such that the sum of all weights is 1. The Softmax operation maps the score to [0,1 ]]Within the range, and a larger score will be larger after normalization, multiplying the attention weight by the value +.>And (5) a matrix, and obtaining the attention output. The attention output is a weighted average of the different positions in the matrix of values, the weights being determined by the attention weights. This enables the network to focus on the part of the input with high weight, the one-dimensional tensor FR13 containing 512 elements; inputting a one-dimensional tensor FR12 and a one-dimensional tensor FR13 into an element-by-element addition layer to obtain an FR14, wherein the one-dimensional tensor FR14 comprises 512 elements; and inputting the one-dimensional tensor FR14 into a seventh full-connection layer and a seventh activation function layer to perform full-connection and activation operation to obtain a one-dimensional tensor FR16 (1), and predicting an input data result through a final Linear function.

In the model training process, the regression prediction measures the loss by using the difference between the regression prediction and the target value, the mean square error is used as a loss function, each Epoch (iteration number) iterates a training set in data, the loss is calculated, the parameter optimization training model is carried out, the learning rate is adjusted at any time, and the batch size is adjusted.

Judging whether the iteration times are smaller than or equal to a set value; if the iteration times are greater than the set value, model training is conducted again; if the iteration number is smaller than or equal to the set value, continuing to judge whether the regression loss is smaller than or equal to a first threshold value; if the regression loss is greater than the first threshold, re-performing model training; and if the regression loss is smaller than or equal to the first threshold value and meets all the set conditions, outputting a final regression model.

In this embodiment, the regression model includes a first full-connection layer, a first activation function layer, a second full-connection layer, a second activation function layer, a third full-connection layer, a third activation function layer, a fourth full-connection layer, a fourth activation function layer, a fifth full-connection layer, a fifth activation function layer, a sixth full-connection layer, a sixth activation function layer, an attention layer one, an element-by-element addition layer, a seventh full-connection layer, and a seventh activation function layer.

Step S42: GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data are respectively input into the classification model for analysis, and a final classification model is obtained.

As shown in fig. 3, step S42 specifically includes:

GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data are respectively and sequentially input into an eighth full-connection layer and an eighth activation function layer to perform full-connection and activation operation, so that one-dimensional tensor FC2 (32,); the one-dimensional tensor FC2 contains 32 elements; inputting the one-dimensional tensor FC2 into a ninth full-connection layer and a ninth activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FC4 (64,); the one-dimensional tensor FC4 contains 64 elements; inputting the one-dimensional tensor FC4 into a tenth full-connection layer and a tenth activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FC6 (128); the one-dimensional tensor FC6 contains 128 elements; inputting the one-dimensional tensor FC6 into an eleventh full-connection layer and an eleventh activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FC8 (256); the one-dimensional tensor FC8 contains 256 elements; inputting the one-dimensional tensor FC6 and the one-dimensional tensor FC8 into the first splicing layer to obtain a one-dimensional tensor FC9 (384,); the one-dimensional tensor FC8 contains 384 elements; inputting a one-dimensional tensor FC8 into a second attention Layer, inheriting the SelfAttention class from Layer to represent a self-defined attention Layer, creating a weight matrix W for calculating attention weights in a building method, performing a process of an attention mechanism in a Call method, performing dot product operation on the input one-dimensional tensor FC8 and the weight matrix W, calculating the attention weights through a Softmax function, and multiplying the attention weights by the input one-dimensional tensor FC8 to obtain a weighted input weighted_F8, namely an output FC10 of the attention mechanism; inputting FC9 and FC10 to the second splicing layer to obtain a one-dimensional tensor FC11 (640,); the one-dimensional tensor FC11 contains 640 elements; inputting the one-dimensional tensor FC11 into a twelfth full-connection layer and a twelfth activation function layer in sequence to perform full-connection and activation operation to obtain a one-dimensional tensor FC13 (512,); the one-dimensional tensor FC13 contains 512 elements; and inputting the one-dimensional tensor FC13 into a thirteenth full-connection layer and a thirteenth activation function layer to perform full-connection and activation operation to obtain a one-dimensional tensor FC15 (category), and performing classification judgment on the input data result through a final Softmax function.

In the model training process, traversing a training set, dividing data into small Batches (Mini-Batches), and inputting current small Batches of data into a model to obtain the prediction output of the model; calculating the loss between the prediction output and the actual label, quantifying the error of the model by using a loss function, and classifying the model by using cross entropy loss; and calculating gradients according to the loss values, updating parameters of the model by using an optimizer, and setting conditions for ending training or conforming to the conditions of the output model. In this embodiment, it is determined whether the iteration number is less than or equal to a preset value; if the iteration times are larger than the preset value, model training is conducted again; if the iteration times are smaller than or equal to a preset value, continuing to judge whether the classification precision is smaller than or equal to a second threshold value; if the classification accuracy is less than or equal to the second threshold value, model training is conducted again; and if the classification precision is greater than the second threshold value and all the set conditions are met, outputting a final classification model.

In this embodiment, the classification model includes an eighth full-connection layer, an eighth activation function layer, a ninth full-connection layer, a ninth activation function layer, a tenth full-connection layer, a tenth activation function layer, an eleventh full-connection layer, an eleventh activation function layer, a first splicing layer, a second attention layer, a second splicing layer, a twelfth full-connection layer, a twelfth activation function layer, a thirteenth full-connection layer, and a thirteenth activation function layer.

In this embodiment, the fault intelligent diagnosis model includes a regression model and a classification model.

Step S5: and inputting the real-time data into a final fault intelligent diagnosis model for diagnosis to obtain a fault evaluation result. The method specifically comprises the following steps:

in this embodiment, the GPF failure data is diagnosed, and the diagnosis result is one of normal, clogging, aging, and damage.

When the diagnosis result is normal, GPF differential pressure, temperature and load signals are all in a normal range, and no abnormality exists; no special handling is required and periodic checks and maintenance are continued to ensure proper operation of the GPF.

When the diagnosis result is that the air flow is blocked, the GPF pressure difference is obviously increased, so that the air flow smoothness is possibly reduced, particles are accumulated in the GPF, and the air flow channel is blocked; the GPF is cleaned using specialized equipment to restore its patency and if the cleaning is not effective, the GPF components may need to be replaced.

When the diagnosis result is aging, the GPF has long service time, the performance can be gradually reduced, and the material performance of the GPF is changed after long-term use and high-temperature environment; the performance of the GPF is checked regularly to ensure that it is still working properly, and if the GPF performance drops significantly, a new GPF component may need to be replaced.

When the diagnosis result is damage, the GPF may be physically damaged, such as cracking, breaking, physical impact, excessive thermal shock, etc., which may cause the GPF to be damaged; checking whether the GPF has physical damage or not, and confirming whether the GPF needs to be replaced or not; if the GPF is severely damaged, the fully new GPF components need to be replaced immediately.

In this embodiment, the fault data of the differential pressure sensor is diagnosed, and the diagnosis result is one of normal, failure, leakage and circuit break.

When the diagnosis result is normal, the differential pressure sensor works normally, outputs accurate differential pressure signals, and the sensor output is stable, accords with the working state of the engine, does not need special treatment, and maintains regular maintenance and inspection.

When the diagnosis result is failure, the differential pressure sensor cannot provide an accurate differential pressure signal, so that the performance of the engine is possibly reduced, and the internal elements of the sensor are failed and connected; the engine performance is poor, the fuel efficiency is reduced, the check connection ensures that the electrical connection of the sensor is normal, no looseness or corrosion is caused, and if the sensor itself is damaged, a new sensor needs to be replaced.

When the diagnosis result is electric leakage, the differential pressure sensor may generate electric leakage, which affects the normal work of the differential pressure sensor, and the problems of wire damage and insulation are solved; the sensor outputs abnormally, and the signal is unstable; checking whether the wires of the sensor are intact, without damage or exposure, ensuring that the insulation of the wires is intact, without wear or damage, and repairing or replacing the wires if any.

When the diagnosis result is that the circuit is broken, the circuit of the differential pressure sensor can be interrupted, so that signals cannot be transmitted, and the electric wires are broken and connected; the sensor cannot output signals, and the performance of the engine is affected; and checking whether the electric wire of the sensor is intact, and ensuring that the sensor is well connected without breakage or loosening, and repairing or replacing the damaged electric wire if necessary.

In this embodiment, the oxygen sensor failure data is diagnosed, and the diagnosis result is one of normal, slow, abnormal output, and damage.

When the diagnosis result is normal, the oxygen sensor can quickly respond to the change of the oxygen content in the tail gas and stably output accurate signals, the combustion adjustment of the engine can quickly respond, the fuel economy and the emission performance are optimally balanced, and the oxygen sensor works normally without special treatment. Periodic checks and maintenance are continued to ensure continued normal operation.

When the diagnosis result is slow, the response speed of the oxygen sensor is slower, the oxygen content of the exhaust gas cannot be timely changed, the engine control system cannot timely adjust the fuel injection quantity, insufficient combustion, excessive emission and reduced fuel economy are possibly caused, the electrical connection of the sensor is ensured to be normal, the sensor is not loosened or corroded, a professional diagnosis tool is used for checking the state of the sensor, whether the sensor needs to be replaced or not is confirmed, and if the response speed of the sensor is too slow, the sensor may need to be replaced by a new sensor which works normally.

When the diagnosis result is abnormal output, the output signal of the oxygen sensor is unstable, fluctuation or jump may occur, the engine control system is interfered by the unstable signal, combustion instability and performance degradation are caused, whether electrical interference or damaged wires exist or not is checked, abnormal output of the sensor may be caused,

If there is an interference source, it is repaired or isolated to ensure that the sensor can function properly, and if the sensor output continues to be abnormal, a new sensor that functions properly may need to be replaced.

When the diagnosis result is that the oxygen sensor is damaged, the output signal of the oxygen sensor is completely invalid, no feedback is provided, the engine control system cannot adjust according to the oxygen content, emission exceeding and fuel economy degradation are possibly caused, a professional diagnosis tool is used for confirming that the oxygen sensor is damaged and cannot be repaired, the damaged sensor is replaced by a brand new sensor which works normally, and necessary calibration and test are carried out after replacement so as to ensure that the sensor works normally.

In this embodiment, the TWC aging data is diagnosed, and the diagnosis is one of normal, aging, poisoning, and failure.

When the diagnosis result is normal, the inlet temperature signal shows that the temperature of the TWC gradually rises to reach the working temperature, the outlet temperature is slightly higher than the inlet temperature, but in the normal range, the conversion efficiency signal is stabilized above a certain level, which indicates that the TWC effectively converts harmful substances into harmless substances, special treatment is not needed, the TWC works normally, and periodic maintenance is continued to ensure the performance.

When the diagnostic result is aging, the inlet temperature gradually increases, but aging may lead to a slow rate of increase, the outlet temperature is slightly higher than the inlet temperature, but the conversion efficiency may decrease, the conversion efficiency signal shows a gradual decrease, indicating that TWC aging affects its conversion performance, considering replacement of TWC to ensure recovery of emissions performance and vehicle performance.

When the diagnostic result is poisoning, the inlet temperature gradually rises, but there may be abnormal temperature fluctuations, the outlet temperature is high relative to the inlet temperature, and may exceed the normal range, the conversion efficiency decreases, indicating that the TWC may be contaminated or poisoned, the fuel mixture is checked, and contaminants in the fuel are removed. The oxygen sensor is checked to ensure that the fuel injection amount is correct and if the problem persists, replacement of the TWC is considered.

When the diagnosis result is failure, the inlet temperature may rise too fast or abnormally, the outlet temperature is obviously raised, the conversion efficiency is extremely low beyond the normal range, the TWC cannot complete effective conversion reaction, and the TWC is replaced to ensure emission control and vehicle performance; at the same time, the fuel and ignition system are checked to prevent similar problems from reoccurring.

In this embodiment, the post-exhaust flow rate data is diagnosed, and the diagnosis result is one of normal, abnormally low, and abnormally high.

And when the diagnosis result is normal, displaying stable rear exhaust flow rate, matching with the engine speed and load, and enabling the relationship between the rear exhaust flow rate and the exhaust system state parameters to be normal. Engine performance and emissions are within acceptable ranges.

When the diagnosis result is abnormally low, the flow rate of the rear exhaust gas is displayed to be obviously lower than the expected flow rate, the problems of exhaust pipe blockage, catalyst damage, choke of the throat pipe and the like can exist, whether the exhaust pipe and the throat pipe are blocked or not is checked, the state of the catalyst is checked, and damaged parts are cleaned or replaced if necessary.

When the diagnosis result is abnormally high, the flow rate of the rear exhaust gas is obviously higher than the expected flow rate, the problems of air leakage of the exhaust system, overheat of the catalyst, failure of the oxygen sensor and the like can exist, whether the air leakage exists in the exhaust system is checked, whether the catalyst is overheated is checked, and whether the oxygen sensor works normally is checked.

In this embodiment, the engine misfire data is diagnosed, and the diagnosis result is one of a normal, single misfire, persistent misfire, and multiple misfire.

And when the diagnosis result is normal, displaying stable engine rotation speed, wherein the crankshaft position signal, the ignition coil signal and the spark plug signal are consistent with expected work, and the engine normally burns the mixed gas without the fire phenomenon. Engine performance and emissions are within normal ranges.

When the diagnosis result is a single fire, the stable engine rotating speed is displayed, the ignition coil of a certain cylinder may not work normally due to the crank shaft position signal, the ignition coil signal and the spark plug signal, the spark plug fault, the ignition coil fault, the ignition system problem and the like, and the single fire may cause the transient decline of the engine power and the increase of the emission; the status of the ignition coil, spark plug and ignition system is checked for necessary maintenance and replacement.

When the diagnosis result is a continuous fire, the unstable engine speed is displayed, a plurality of cylinders may have ignition problems due to crank shaft position signals, ignition coil signals and spark plug signals, the ignition system is in fault, the fuel supply is insufficient, the cylinder pressure is abnormal and the like, and the continuous fire leads to unstable operation of the engine, power reduction and obvious increase of emission; the entire ignition system and fuel supply system are inspected, problems are found and maintenance is performed.

When the diagnosis result is that a plurality of fires happen, unstable engine rotating speed is displayed, frequent fire phenomena can occur to a plurality of cylinders through a crankshaft position signal, an ignition coil signal and a spark plug signal, and serious ignition system problems and fuel system faults are caused. The engine cannot work normally due to multiple fire, and can be stopped or cannot be started; the entire ignition system, fuel system and cylinder pressure are immediately checked for emergency maintenance.

In this embodiment, the exhaust pipe vibration data is diagnosed, and the diagnosis result is one of normal, abnormal vibration caused by engine vibration, abnormal vibration caused by exhaust gas pulse, and abnormal vibration caused by road surface bump.

When the diagnosis result is normal, the stable vibration acceleration is displayed and matched with the engine speed and working condition, the engine speed signal and the tail gas pressure signal are consistent with the expected work, the vibration of the exhaust pipe is in a normal range, no abnormal vibration phenomenon exists, and the problems of noise and structural damage are small.

When the diagnosis result is abnormal vibration caused by engine vibration, abnormal high vibration acceleration is displayed, the frequency can be consistent with the engine rotating speed, an engine rotating speed signal and a tail gas pressure signal can be consistent with the actual working condition, the problems of unbalanced engine, unstable ignition and the like can cause abnormal vibration, the abnormal vibration can cause obvious increase of noise, the durability of an exhaust system can be influenced, the balance of the engine is checked, whether the ignition system works normally is checked, and necessary adjustment and maintenance are carried out.

When the diagnosis result is abnormal vibration caused by exhaust gas pulse, the display frequency is consistent with the vibration of the exhaust gas pulse frequency, the engine rotating speed signal and the exhaust gas pressure signal may be consistent with the actual working condition, the exhaust gas pulse vibration may be caused by the problems of improper design of an exhaust system, damage of an exhaust pipe corrugated pipe and the like, the abnormal vibration caused by the exhaust gas pulse may cause the increase of vibration noise, the structure of the exhaust system is adversely affected, the design of the exhaust system is checked, the components such as the corrugated pipe and the like are ensured not to be damaged, and the exhaust gas pulse is replaced if necessary.

When the diagnosis result is abnormal vibration caused by road surface bump, unstable vibration acceleration is displayed, the frequency can be consistent with the road surface bump frequency, an engine rotating speed signal and a tail gas pressure signal can be consistent with the actual working condition, the abnormal vibration can be caused by the problems of unstable exhaust pipe suspension system, damaged supporting materials and the like, the abnormal vibration caused by road surface bump can cause uneven stress of the exhaust pipe, the accelerating structure is damaged, the exhaust pipe suspension system is inspected, the supporting materials and the hanging parts are ensured to be normal, and necessary maintenance and replacement are carried out.

Example 2

As shown in fig. 4, the present invention discloses a light vehicle exhaust fault diagnosis system, the system comprising:

the state parameter acquisition module 10 is configured to acquire a state parameter.

The signal processing module 20 is configured to perform a signal processing operation on the state parameter to obtain a standard signal parameter.

The signal parameter combination module 30 is configured to combine the standard signal parameters to obtain fault diagnosis data.

The fault intelligent diagnosis model identification module 40 is configured to input fault diagnosis data into the fault intelligent diagnosis model for analysis, so as to obtain a final fault intelligent diagnosis model.

The test evaluation module 50 is used for inputting real-time data into the final fault intelligent diagnosis model for diagnosis, and obtaining a fault evaluation result.

As an alternative embodiment, the fault intelligent diagnosis model identification module 40 of the present invention specifically includes:

the regression model identification sub-module 401 is configured to input the oxygen sensor fault data, the TWC aging data, and the post-exhaust flow rate data into the regression model for analysis, respectively, to obtain a final regression model.

The classification model identification sub-module 402 is configured to input GPF fault data, differential pressure sensor fault data, engine misfire data, and exhaust pipe vibration data into the classification model respectively for analysis, to obtain a final classification model.

As an alternative embodiment, the regression model identification sub-module 401 of the present invention specifically includes:

the regression model building unit is used for inputting the oxygen sensor fault data, the TWC aging data and the rear exhaust flow velocity data into the first full-connection layer, the first activation function layer, the second full-connection layer, the second activation function layer, the third full-connection layer, the third activation function layer, the fourth full-connection layer, the fourth activation function layer, the fifth full-connection layer, the fifth activation function layer, the sixth full-connection layer, the sixth activation function layer, the attention layer one, the element-by-element addition layer, the seventh full-connection layer and the seventh activation function layer in sequence respectively for analysis, so that a final regression model is obtained.

As an alternative implementation manner, the classification model identification sub-module 402 specifically comprises:

the classification model building unit is used for inputting GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data into the eighth full-connection layer, the eighth activation function layer, the ninth full-connection layer, the ninth activation function layer, the tenth full-connection layer, the tenth activation function layer, the eleventh full-connection layer, the eleventh activation function layer, the first splicing layer, the second attention layer, the second splicing layer, the twelfth full-connection layer, the twelfth activation function layer, the thirteenth full-connection layer and the thirteenth activation function layer in sequence respectively for analysis, and obtaining a final classification model.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. A method for diagnosing an exhaust failure of a light vehicle, the method comprising:

step S1: acquiring state parameters;

Step S2: performing signal processing operation on the state parameters to obtain standard signal parameters;

step S3: combining the standard signal parameters to obtain fault diagnosis data;

step S4: inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model;

step S5: and inputting the real-time data into a final fault intelligent diagnosis model for diagnosis to obtain a fault evaluation result.

2. The method for diagnosing an exhaust gas failure of a light vehicle according to claim 1, wherein said obtaining state parameters specifically includes:

the state parameters include a GPF differential pressure signal, a GPF temperature signal, a GPF load signal, a differential pressure sensor signal, an engine speed signal, a throttle opening signal, an air-fuel ratio sensor signal, a post-oxygen sensor signal, a fuel correction signal, a TWC inlet temperature signal, a TWC outlet temperature signal, a TWC conversion efficiency signal, a post exhaust flow rate signal, a load signal, an ignition coil signal, a spark plug signal, a crankshaft position signal, an exhaust gas pressure signal, and an exhaust pipe vibration acceleration signal.

3. The method for diagnosing exhaust gas failure of a light vehicle according to claim 1, wherein said combining said standard signal parameters to obtain failure diagnosis data comprises:

The fault diagnosis data comprises GPF fault data, differential pressure sensor fault data, oxygen sensor data, TWC aging data, post exhaust flow rate data, engine misfire data and exhaust pipe vibration data;

the GPF fault data comprises a standard GPF differential pressure signal, a standard GPF temperature signal and a standard GPF load signal; the differential pressure sensor fault data comprises a standard differential pressure sensor signal, a standard engine speed signal and a standard throttle opening signal; the oxygen sensor data includes a standard air-fuel ratio sensor signal, a standard post-oxygen sensor signal, a standard fuel correction signal, and a standard engine speed signal; the TWC aging data includes a standard TWC inlet temperature signal, a standard TWC outlet temperature signal, and a standard TWC conversion efficiency signal; the rear exhaust flow rate data includes a standard rear exhaust flow rate signal, a standard load signal, and a standard engine speed signal; the engine misfire data includes a standard ignition coil signal, a standard spark plug signal, a standard crankshaft position signal, and a standard engine speed signal; the exhaust pipe vibration data comprises a standard tail gas pressure signal, a standard exhaust pipe vibration acceleration signal and a standard engine speed signal.

4. The exhaust gas fault diagnosis method for a light vehicle according to claim 1, wherein the step of inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model comprises the following steps:

the fault intelligent diagnosis model comprises a regression model and a classification model;

respectively inputting oxygen sensor fault data, TWC aging data and rear exhaust gas flow rate data into the regression model for analysis to obtain a final regression model;

and respectively inputting GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data into the classification model for analysis to obtain a final classification model.

5. The method for diagnosing an exhaust gas failure of a light vehicle according to claim 4, wherein the steps of inputting oxygen sensor failure data, TWC aging data and post exhaust gas flow rate data into the regression model respectively for analysis, and obtaining a final regression model include:

the regression model comprises a first full-connection layer, a second activation function layer, a third full-connection layer, a third activation function layer, a fourth full-connection layer, a fourth activation function layer, a fifth full-connection layer, a fifth activation function layer, a sixth full-connection layer, a sixth activation function layer, a first attention layer, an element-by-element addition layer, a seventh full-connection layer and a seventh activation function layer;

And respectively and sequentially inputting oxygen sensor fault data, TWC aging data and post-exhaust flow velocity data into the first full-connection layer, the first activation function layer, the second full-connection layer, the second activation function layer, the third full-connection layer, the third activation function layer, the fourth full-connection layer, the fourth activation function layer, the fifth full-connection layer, the fifth activation function layer, the sixth full-connection layer, the sixth activation function layer, the attention layer I, the element-by-element addition layer, the seventh full-connection layer and the seventh activation function layer to analyze, thereby obtaining a final regression model.

6. The method for diagnosing an exhaust gas failure of a light vehicle according to claim 4, wherein the steps of inputting GPF failure data, differential pressure sensor failure data, engine misfire data, and exhaust pipe vibration data into the classification model respectively for analysis, and obtaining a final classification model include:

the classification model comprises an eighth full-connection layer, an eighth activation function layer, a ninth full-connection layer, a ninth activation function layer, a tenth full-connection layer, a tenth activation function layer, an eleventh full-connection layer, an eleventh activation function layer, a first splicing layer, a second attention layer, a second splicing layer, a twelfth full-connection layer, a twelfth activation function layer, a thirteenth full-connection layer and a thirteenth activation function layer;

GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data are sequentially input into the eighth full-connection layer, the eighth activation function layer, the ninth full-connection layer, the ninth activation function layer, the tenth full-connection layer, the tenth activation function layer, the eleventh full-connection layer, the eleventh activation function layer, the first splicing layer, the second attention layer, the second splicing layer, the twelfth full-connection layer, the twelfth activation function layer, the thirteenth full-connection layer and the thirteenth activation function layer respectively, and analysis is carried out to obtain a final classification model.

7. A light-duty vehicle exhaust fault diagnosis system, the system comprising:

the state parameter acquisition module is used for acquiring state parameters;

the signal processing module is used for performing signal processing operation on the state parameters to obtain standard signal parameters;

the signal parameter combination module is used for combining the standard signal parameters to obtain fault diagnosis data;

the fault intelligent diagnosis model identification module is used for inputting the fault diagnosis data into a fault intelligent diagnosis model for analysis to obtain a final fault intelligent diagnosis model;

And the test evaluation module is used for inputting the real-time data into the final fault intelligent diagnosis model for diagnosis to obtain a fault evaluation result.

8. The exhaust gas failure diagnosis system of a light vehicle according to claim 7, wherein the failure intelligent diagnosis model identification module specifically comprises:

the regression model identification submodule is used for respectively inputting the oxygen sensor fault data, the TWC aging data and the rear exhaust flow velocity data into a regression model for analysis to obtain a final regression model;

and the classification model identification sub-module is used for respectively inputting the GPF fault data, the differential pressure sensor fault data, the engine fire data and the exhaust pipe vibration data into the classification model for analysis to obtain a final classification model.

9. The exhaust gas failure diagnosis system of a light vehicle according to claim 8, wherein the regression model identification sub-module specifically comprises:

the regression model building unit is used for inputting the oxygen sensor fault data, the TWC aging data and the rear exhaust flow velocity data into the first full-connection layer, the first activation function layer, the second full-connection layer, the second activation function layer, the third full-connection layer, the third activation function layer, the fourth full-connection layer, the fourth activation function layer, the fifth full-connection layer, the fifth activation function layer, the sixth full-connection layer, the sixth activation function layer, the attention layer one, the element-by-element addition layer, the seventh full-connection layer and the seventh activation function layer in sequence respectively for analysis, so that a final regression model is obtained.

10. The exhaust gas failure diagnosis system of a light vehicle according to claim 8, wherein the classification model identification sub-module specifically comprises:

the classification model building unit is used for inputting GPF fault data, differential pressure sensor fault data, engine fire data and exhaust pipe vibration data into the eighth full-connection layer, the eighth activation function layer, the ninth full-connection layer, the ninth activation function layer, the tenth full-connection layer, the tenth activation function layer, the eleventh full-connection layer, the eleventh activation function layer, the first splicing layer, the second attention layer, the second splicing layer, the twelfth full-connection layer, the twelfth activation function layer, the thirteenth full-connection layer and the thirteenth activation function layer in sequence respectively for analysis, and obtaining a final classification model.