A method and device for smelling a car

Abstract

The invention discloses a method and device for automobile olfactory identification, and relates to the technical field of automobile gas olfactory identification detection. The olfactory identification method comprises preprocessing, sparse dictionary construction, gas identification and postprocessing; a plurality of sensors of different types are used to capture the multidimensional characteristics of automobile exhaust gas, and environmental parameters are recorded at the same time, and a data set is obtained by preprocessing; the response pattern of a known gas is used as an initial dictionary atom, and as new data is collected, the response pattern of the new gas is added to the dictionary through an online learning algorithm, and for the new sensor response, the best sparse representation is found in an adaptive dictionary; according to the result of sparse coding, the most relevant dictionary atom is identified; based on the identified gas composition and the sensor response intensity, the concentration of each gas is estimated, and the odor intensity of the detected gas is analyzed and identified in combination with the automobile emission standard and the identified gas concentration, so as to better deal with the problem of interference caused by non-monitored gases in a complex gas environment.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile gas olfactory detection, and in particular to a method and device for automobile olfactory detection. Background Art

[0002] The core of the artificial olfactory system includes a data acquisition module and a computing unit module. The computing unit converts the instrument response value and the predicted value. After completing the fitting calculation, it finally converts the electrical signal intensity into an observation value with actual physical meaning to complete the instrument detection task. In this process, the sensor part of the data acquisition module is the key core part that converts the gas information to be tested into the instrument response.

[0003] In the field of automobile inspection, traditionally, the monitoring of the internal environment of automobiles mainly relies on manual sampling and inspection. This method has the disadvantages of low efficiency, high cost, and untimely data collection. With the development of the Internet of Things and sensor technology, more and more intelligent monitoring systems are introduced into modern automobile production. The in-car odor intensity olfactory instrument used by the intelligent monitoring system is dedicated to the in-car environment. The gas sensor converts the gas to be tested into an electrical signal and processes it, analyzes and identifies the odor intensity of the gas to be tested, and realizes real-time monitoring and data collection of the internal environment of the car.

[0004] In actual application, the sensor or sensor array in the olfactory detector will be affected by the ambient temperature and humidity, the interfering gas composition in the environment and its own operation drift, resulting in nonlinear deviation between the response value and the true value. For the interference of non-monitored gases, the traditional solution is to reduce the interference of abnormal gases on the data collection side as much as possible. However, in the actual application environment, the type and content of gases are highly uncertain. The use of sensor arrays with stronger anti-interference capabilities will increase the cost of application. On the other hand, from the perspective of the actual gas distribution environment, the gas content in the environment is correlated. Taking NMHC gas as an example, it is a mixed hydrocarbon gas produced in the chemical reaction of organic matter. When organic matter is produced, CO, CH4, etc. in the environment will also change. The content of various gases will eventually reach a dynamic balance in the air. Therefore, such a complex gas environment is the normal working state of the artificial olfactory system.

[0005] Therefore, a method and device for car smelling are proposed to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to solve the problem that the prior art is difficult to detect in automobiles and difficult to deal with complex gas environments, and to propose a method and device for detecting automobiles.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A vehicle olfactory identification method, including preprocessing, sparse dictionary construction, gas identification and postprocessing;

[0009] Use multiple sensors of different types to capture the multi-dimensional characteristics of the gases generated by the car, and record environmental parameters at the same time to correct their impact on the sensor response and preprocess the data set; use the response patterns of known gases as the initial dictionary atoms, and as new data is collected, add the response patterns of new gases to the dictionary through an online learning algorithm, and find the best sparse representation in the adaptive dictionary for new sensor responses; identify the most relevant dictionary atoms based on the results of sparse coding; estimate the concentration of each gas based on the identified gas composition and sensor response intensity for in-vehicle environment analysis, and analyze and identify the odor intensity of the detected gas.

[0010] Further, the preprocessing step includes noise removal and normalization operations, wherein the wavelet transform decomposes the signal into components of different frequencies, the high-frequency components represent the noise, and the sensor response matrix is ​​decomposed using a two-dimensional wavelet transform, and the matrix is ​​decomposed into low-frequency and high-frequency parts based on a two-dimensional discrete wavelet transform;

[0011] ;

[0012] A: Filtering in the horizontal and vertical directions is achieved through convolution operations, where the low-pass filter h and the high-pass filter g are used to extract approximate and detail information;

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] Among them, * represents the convolution operation, and Respectively represent the transpose of the filter;

[0018] B: After getting each Approx and Detail, downsample them to reduce the size of the matrix;

[0019] ;

[0020] ;

[0021] ;

[0022] ;

[0023] in, It means removing half of the sampling points in each row and column;

[0024] C: Repeat steps A and B step by step to further decompose the low-frequency subbands to form a multi-layer decomposition structure and obtain approximate and detailed information at different scales;

[0025] In the process of wavelet transform, the noise is concentrated in the high-frequency detail part. The noise is removed by setting the wavelet coefficient threshold of the high-frequency detail part to zero.

[0026] Furthermore, the environmental parameters are corrected, and the sensor response is corrected by a linear regression model, and the model is expressed as:

[0027] ;

[0028] W is the correction coefficient matrix. The goal of matrix W is to correct the sensor response through environmental parameters. The correction coefficient matrix W is estimated using the least squares method, that is, minimizing the following loss function:

[0029] ;

[0030] in, Represents the Frobenius norm, that is, there is a loss function:

[0031] ;

[0032] Calculate the optimal solution:

[0033] ;

[0034] In this process, it is necessary to keep the matrix is a reversible matrix, otherwise it needs to be regularized. By solving the above equation, we can get the correction coefficient matrix The analytical solution of

[0035] ;

[0036] Furthermore, the sensor response is corrected using a nonlinear regression model, and the RBF model used is expressed as:

[0037] ;

[0038] ;

[0039] in, is the number of RBFs, representing the complexity of the model, It is The weight of each RBF, Used to indicate the The output of an RBF, is the center, is the width parameter;

[0040] Estimate model parameters by minimizing the loss function;

[0041] ;

[0042] Furthermore, constructing an adaptive sparse coding dictionary includes:

[0043] S1: Select a set of initial dictionary atoms, use the response patterns of known gases as the initial dictionary atoms, each vector represents the response pattern of a gas, and these atoms form the initial dictionary matrix D0;

[0044] S2: The response to the newly collected gas is expressed as , generating a new sensor response sample ;

[0045] ;

[0046] The new sensor response sample Bring it in and update the dictionary using the online learning algorithm:

[0047] ;

[0048] in, Represents the learning rate, which is used to control the step size of dictionary update;

[0049] ;

[0050] for Coefficient encoding, is a regularization parameter used to balance reconstruction error and sparsity to obtain a sparse representation matrix , used to update the dictionary, represents the L2 norm (i.e., Euclidean norm), which is used to measure the reconstruction error; Represents the L1 norm, which is used to promote sparsity.

[0051] Furthermore, the iterative process is as follows:

[0052] For each sample :

[0053] A: Use the least squares form of the LASSO problem to update Initialize the matrix to a zero vector, or Use the sparse representation from the previous step as the initial value;

[0054] ;

[0055] B: Yes Perform soft threshold operation, that is, for each component Perform threshold processing;

[0056] ;

[0057] C: Update residual ;

[0058] ;

[0059] Using residuals Update dictionary matrix;

[0060] ;

[0061] Determine whether the stop condition is met. If so, end the iteration, otherwise return to continue the iteration; continuously update the sparse representation coefficients and dictionary matrix D0 until convergence to the optimal solution or reaching a predetermined number of iterations, the optimal judgment criterion can be based on one or more of the judgment conditions such as the residual threshold, the sparse representation matrix change threshold, the number of iterations, etc.

[0062] An automobile olfactory identification device includes a sensor array, an environmental parameter sensor, a data acquisition device, a computing unit, a user interface, a storage device, and a communication module;

[0063] Among them, the sensor array is used to capture the multi-dimensional characteristics of the gas in the vehicle. The gas sensor includes sensors for detecting various gas components, and the environmental parameter sensor is used to record environmental parameters. The data acquisition device is used to receive and process the data of the sensor array and the environmental parameter sensor, and perform preprocessing and feature extraction. The computing unit is used to execute the algorithm, including sparse representation, gas identification, concentration estimation, fault diagnosis and other calculation steps. The user interface is used to display the identification results, concentration estimation and fault diagnosis information, as well as to provide decision support suggestions. The storage device is used to store sensor data, model parameters and historical records for performance evaluation and model optimization. The communication module is used for data exchange and remote control with other systems or the cloud.

[0064] Compared with the prior art, the present invention has the following advantages:

[0065] The present invention uses a multi-sensor array to capture the multi-dimensional characteristics of the gas in the vehicle, which can more comprehensively analyze the gas composition, and correct the sensor response in combination with the data recorded by the environmental parameter sensor, thereby improving the accuracy and stability of gas identification.

[0066] By building an adaptive sparse coding dictionary, it is possible to better handle the problem of non-monitored gas interference in complex gas environments, and improve the accuracy and robustness of gas identification. Taking into account multiple factors such as sensor response, environmental parameters, sparse representation and gas identification, the system's comprehensive performance and practicality are improved. It can not only identify gas components, but also estimate the concentration of each gas, analyze the odor intensity of the detected gas, provide maintenance suggestions, optimize emission strategies and other decision support information, and help users better manage and maintain the gas environment in the car. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a method for car smelling proposed by the present invention. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0069] An automobile olfactory identification device includes a sensor array, an environmental parameter sensor, a data acquisition device, a computing unit, a user interface, a storage device, and a communication module;

[0070] The sensor array is used to capture the multi-dimensional characteristics of the gas inside the car, and is mainly used to monitor the air environment inside the car during the production and delivery process of the car. The sensor should include sensors that can detect various gas components, such as volatile organic compounds (VOC), nitrogen oxides (NO x ), carbon monoxide (CO), etc.; environmental parameter sensors are used to record environmental parameters such as temperature, humidity, air pressure, etc. to correct the influence of sensor response; data acquisition equipment is used to receive and process data from sensor arrays and environmental parameter sensors, and perform preprocessing and feature extraction; the computing unit is used to execute algorithms, including sparse representation, gas identification, concentration estimation, fault diagnosis and other calculation steps; the user interface is used to display identification results, concentration estimation and fault diagnosis information, and to provide decision support recommendations; the storage device is used to store sensor data, model parameters and historical records for performance evaluation and model optimization; the communication module is used to exchange data and remotely control with other systems or the cloud.

[0071] When monitoring the internal gas of a vehicle, the device can more accurately reflect the actual content of various gases in the internal environment of the vehicle. The sensor array therein can better adapt to complex gas environments, including the high uncertainty of gas types and contents and the correlation between gas contents, thereby more effectively coping with actual application environments; the sensor array and data processing methods are used to cope with the nonlinear deviation of environmental factors on sensor response, thereby improving the monitoring accuracy and stability of the sensor array and making it better suitable for complex gas environments.

[0072] Reference Figure 1 , a method for car olfactory identification, including preprocessing, sparse dictionary construction, gas identification and postprocessing, is applicable to the above device, and the specific method is as follows;

[0073] 1. New data collection and data preprocessing: 1.1

[0075] Collect raw data from the sensor array, including multi-dimensional characteristics of the gas in the car and environmental parameters (such as temperature, humidity, etc.), sensor response vector representation and environmental parameter vector representation, the sensor array receives the signal, represented as a response vector X (M sensor groups);

[0076] ;

[0077] P environmental parameters: temperature, humidity, and air pressure;

[0078] ; 1.2

[0080] In the preprocessing stage, the sensor data is processed to remove noise to improve data quality and accuracy, and the sensor data is normalized to the same scale range to eliminate the dimensional differences between different sensors and facilitate subsequent processing.

[0081] Using wavelet transform for denoising, the initial data is represented as X0, E0, that is;

[0082] ;

[0083] ;

[0084] Wavelet transform decomposes the signal into components of different frequencies, where the high-frequency components represent noise. The sensor response matrix is ​​decomposed using two-dimensional wavelet transform. Based on two-dimensional discrete wavelet transform, the matrix is ​​decomposed into low-frequency and high-frequency parts;

[0085] ;

[0086] A: Filtering in the horizontal and vertical directions is achieved through convolution operations, where the low-pass filter h and high-pass filter g are used to extract approximate and detail information;

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] Among them, * represents the convolution operation, and Respectively represent the transpose of the filter;

[0092] B: After obtaining each subband (Approx and Detail), downsample it to reduce the size of the matrix;

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] in, It means removing half of the sampling points in each row and column;

[0098] C: Repeat steps A and B step by step to further decompose the low-frequency subbands to form a multi-layer decomposition structure and obtain approximate and detailed information at different scales;

[0099] In the process of wavelet transform, the noise is concentrated in the high-frequency detail part. The noise is removed by setting the wavelet coefficient threshold of the high-frequency detail part to zero.

[0100] The normalization operation normalizes the sensor data to the same scale range to eliminate the dimensional differences between different sensors and facilitate subsequent processing;

[0101] Sensor response matrix X:

[0102] ;

[0103] Calculate the mean:

[0104] ;

[0105] in, Represents the i-th column of the matrix, i=1,2,...,M;

[0106] Calculate the standard deviation:

[0107] ;

[0108] Finally, the normalized sensor response matrix is ​​calculated, where the data in each column has zero mean and unit variance;

[0109] ; 1.3

[0111] Environmental parameter correction, using a linear regression model to correct the sensor response, the model is expressed as:

[0112] ;

[0113] W is the correction coefficient matrix. The goal of matrix W is to correct the sensor response through environmental parameters. The correction coefficient matrix W is estimated using the least squares method, that is, minimizing the following loss function:

[0114] ;

[0115] in, Represents the Frobenius norm, that is, there is a loss function:

[0116] ;

[0117] Calculate the optimal solution:

[0118] ;

[0119] In this process, it is necessary to keep the matrix is a reversible matrix, otherwise it needs to be regularized. By solving the above equation, we can get the correction coefficient matrix The analytical solution of

[0120] ;

[0121] The sensor response is corrected using a nonlinear regression model, and the RBF model used is expressed as:

[0122] ;

[0123] ;

[0124] in, is the number of RBFs, representing the complexity of the model, It is The weight of each RBF, Used to indicate the The output of an RBF, is the center, is the width parameter;

[0125] Similarly, the model parameters are estimated by minimizing the loss function;

[0126] ;

[0127] deal with Then get the response vector ,The preprocessing steps can help improve the quality and accuracy of sensor data, and provide a more reliable data basis for subsequent gas identification and ,concentration estimation.

[0128] 2. Constructing an adaptive sparse coding dictionary: 2.1

[0130] Select a set of initial dictionary atoms, use the response patterns of known gases as the initial dictionary atoms, each vector represents the response pattern of a gas, and these atoms form the initial dictionary matrix D0;

[0131] ;

[0132] ; 2.2

[0134] The response to the newly collected gas is expressed as , generating a new sensor response sample ;

[0135] ;

[0136] The new sensor response sample Bring it in and update the dictionary using the online learning algorithm:

[0137] ;

[0138] in, Represents the learning rate, which is used to control the step size of dictionary update;

[0139] ;

[0140] for Coefficient encoding, is a regularization parameter used to balance reconstruction error and sparsity to obtain a sparse representation matrix , used to update the dictionary, represents the L2 norm (i.e., Euclidean norm), which is used to measure the reconstruction error; Represents the L1 norm, which is used to promote sparsity.

[0141] The iteration process is as follows:

[0142] For each sample :

[0143] A: Use the least squares form of the LASSO problem to update Initialize the matrix to a zero vector, or Use the sparse representation from the previous step as the initial value;

[0144] ;

[0145] B: Yes Perform soft thresholding operation, that is, for each component Perform threshold processing;

[0146] ;

[0147] C: Update residual ;

[0148] ;

[0149] Using residuals Update dictionary matrix;

[0150] ;

[0151] Determine whether the stop condition is met. If so, end the iteration, otherwise return to continue the iteration; continuously update the sparse representation coefficients and dictionary matrix D0 until convergence to the optimal solution or reaching a predetermined number of iterations, the optimal judgment criterion can be based on one or more of the judgment conditions such as the residual threshold, the sparse representation matrix change threshold, the number of iterations, etc.

[0152] The least squares form of the LASSO method is suitable for controlling sparsity and is suitable for the sparsity requirements in sparse coding. Compared with the general co-iterative soft threshold algorithm, although the co-iterative soft threshold can also be used to update the dictionary, it is usually more suitable for solving general sparse representation problems rather than the sparsity constraints in LASSO. Therefore, when constructing an adaptive sparse coding dictionary to deal with the problem of non-monitored gas interference in a complex gas environment, it is more appropriate to use the least squares form of the LASSO problem to update the dictionary.

[0153] 3. Gas Identification

[0154] According to the results of sparse coding , identify the most relevant dictionary atoms, that is, the most likely gas components, and the threshold for sparse coding is , then the identified gas components can be expressed as

[0155] ;

[0156] in, is a sparse representation matrix vector The i-th vector of , where i is the index of the dictionary atom, represents the i-th gas component.

[0157] 4. Post-processing and decision support

[0158] ;

[0159] Where C represents the concentration of the gas corresponding to GasID, and f is the concentration estimation function. Based on the concentration estimation results and fault diagnosis information, the system provides optimization suggestions to users, such as adjusting engine parameters, replacing parts, decorative materials, or performing repairs.

[0160] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for smelling a car, characterized in that: Includes preprocessing, sparse dictionary construction, gas identification and postprocessing; Use multiple sensors of different types to capture the multi-dimensional characteristics of the gases produced by the car, and record environmental parameters to correct their effects on sensor response and pre-process the data set; use the response patterns of known gases as initial dictionary atoms, and as new data is collected, add the response patterns of new gases to the dictionary through an online learning algorithm, and find the best sparse representation in the adaptive dictionary for new sensor responses; According to the results of sparse coding, the most relevant dictionary atoms are identified; based on the identified gas components and sensor response intensity, the concentration of each gas is estimated, and combined with the automobile emission standards and the identified gas concentration, it is used for in-vehicle environment analysis to analyze and identify the odor intensity of the detected gas; The preprocessing steps include noise removal and normalization operations, using wavelet transform to decompose the signal into components of different frequencies, with high-frequency components representing noise, and using two-dimensional wavelet transform to decompose the sensor response matrix X. Based on two-dimensional discrete wavelet transform, the matrix X is decomposed into low-frequency and high-frequency parts; ； The steps to obtain DWT(X) include A, B and C steps: A: Filtering in the horizontal and vertical directions is achieved through convolution operations, where the low-pass filter h and the high-pass filter g are used to extract approximate and detail information; ； ； ； ； Among them, * represents the convolution operation, and Respectively represent the transpose of the filter; B: After getting each Approx and Detail, downsample them to reduce the size of the matrix; ； ； ； ； in, It means removing half of the sampling points in each row and column; C: Repeat steps A and B step by step to further decompose the low-frequency subbands to form a multi-layer decomposition structure and obtain approximate and detailed information at different scales; In the process of wavelet transform, the noise is concentrated in the high-frequency detail part, and the noise is removed by setting the wavelet coefficient threshold of the high-frequency detail part to zero; Among them, building an adaptive sparse coding dictionary includes the following steps: S1: Select a set of initial dictionary atoms, use the response patterns of known gases as the initial dictionary atoms, each vector represents the response pattern of a gas, and these atoms form the initial dictionary matrix D0; ； S2: The response to the newly collected gas is expressed as , generating a new sensor response sample ; ； The new sensor response sample Bring it in and update the dictionary using the online learning algorithm: ； Where D t Represents the current dictionary matrix, D t+1 Indicates D t The new dictionary matrix obtained after one update, D t Through multiple iterations of D0, Represents the learning rate, which is used to control the step size of dictionary update. is the sparse representation matrix used when updating Dt, yes The transposed matrix of ； right Coefficient encoding, is a regularization parameter used to balance reconstruction error and sparsity to obtain a sparse representation matrix , used to update the dictionary, represents the L2 norm, which is used to measure the reconstruction error; Represents the L1 norm, which is used to promote sparsity.

2. A method for smelling a car according to claim 1, characterized in that: The sensor response matrix is ​​corrected for environmental parameters, and the sensor response is corrected using a linear regression model. The model is expressed as: ； Where X is the sensor response matrix, is the information matrix after environmental parameter correction, is the sensor response matrix obtained after normalization of the original sensor response matrix, E is the environmental parameter matrix, W is the correction coefficient matrix, the goal of the matrix W is to correct the sensor response through environmental parameters, and the correction coefficient matrix W is estimated using the least squares method to minimize the following loss function: ； in, represents the Frobenius norm, and the loss function is obtained : ； Calculating the loss function Optimal solution: ； When calculating the loss function In the process of optimal solution, it is necessary to keep the matrix is a reversible matrix, otherwise it needs to be regularized by solving the above loss function Optimal solution equation, get the correction coefficient matrix The analytical solution of .

3. A method for smelling a car according to claim 2, characterized in that: The sensor response is corrected using a nonlinear regression model, and the RBF model used is expressed as: ； ； in, is the number of RBFs, representing the complexity of the model. It is The weight of each RBF, Used to indicate the The output of an RBF, is the center, is the width parameter; Estimate model parameters by minimizing the loss function; 。 4. A method for smelling a car according to claim 1, characterized in that: The dictionary is updated through the online learning algorithm. The iterative process is as follows: For each sample : A: Use the least squares form of the LASSO problem to update Initialize the matrix to a zero vector, or Use the sparse representation from the previous step as the initial value; ； B: Yes Perform soft threshold operation, that is, for each component Perform threshold processing; ； C: Update residual ; ； Using residuals Update dictionary matrix; ； Determine whether the stop condition is met. If so, end the iteration, otherwise return to continue the iteration; continuously update the sparse representation coefficients and dictionary matrix until convergence to the optimal solution or reaching a predetermined number of iterations, the optimal judgment criterion can be based on one or more of the judgment conditions such as the residual threshold, the sparse representation matrix change threshold, the number of iterations, etc.

5. A vehicle sniffing device for implementing the vehicle sniffing method according to any one of claims 1 to 4, comprising a sensor array, an environmental parameter sensor, a data acquisition device, a computing unit, a user interface, a storage device and a communication module; in, The sensor array is used to capture the multi-dimensional characteristics of the gas in the car. The sensors should include sensors that can detect various gas components, and the environmental parameter sensors are used to record environmental parameters; the data acquisition device is used to receive and process the data of the sensor array and the environmental parameter sensor, and perform preprocessing and feature extraction; the computing unit is used to execute the algorithm, including sparse representation, gas identification, concentration estimation, fault diagnosis and other calculation steps; the user interface is used to display the identification results, concentration estimation and fault diagnosis information, and provide suggestions for decision support; Storage devices are used to store sensor data, model parameters, and historical records for performance evaluation and model optimization; The communication module is used for data exchange and remote control with other systems or the cloud.

Images