Automobile oxygen sensor production line automatic test and inspection system based on cloud platform

Abstract

The invention relates to the technical field of automobile electronic manufacturing, in particular to an automobile oxygen sensor production line automatic test and inspection system based on a cloud platform. The system comprises a data acquisition module, a phase-space reconstruction module, a manifold analysis module and a feedback control module. The system collects parameters such as sensor voltage and heating rod current, and constructs a dimensionless phase space state vector; the core of the method is to calculate a difference integral between a vector actual evolution speed and a good product manifold field, and generate an accumulated trajectory deviation score to judge quality. According to the method, transient high-frequency data which does not reach heat balance is utilized, judgment depending on a long-time steady state is converted into short-time transient analysis, the problem that the production takt is slow due to the fact that heat balance needs to be waited is solved, and the detection efficiency is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of automotive electronic component manufacturing and testing technology, specifically to an automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform. Background Technology

[0002] With the continuous development of the automotive industry, higher requirements are being placed on the production efficiency and testing quality of core components such as oxygen sensors. The automated testing process of oxygen sensor production lines faces severe challenges, especially in balancing testing speed and accuracy. Currently, existing oxygen sensor testing technologies generally rely on physical thermal equilibrium. During testing, data can only be collected and quality determined after the sensor has fully heated up and reached a steady state. However, this traditional steady-state testing method results in excessively long testing cycles for individual products, severely slowing down the overall production pace. Furthermore, during the transient phase when the sensor starts heating, physical quantities such as voltage, current, and internal resistance exhibit highly nonlinear and chaotic characteristics. Traditional linear threshold determination methods are unable to analyze these complex data patterns and cannot effectively identify product defects without waiting for thermal equilibrium. Therefore, how to solve the technical problem of slow production cycle caused by waiting for physical thermal equilibrium during the testing process, and how to overcome the problem that transient data chaos is difficult to judge through linear thresholds, have become urgent problems to be solved in this field. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform. Specifically, the technical solution of this invention includes: The data acquisition module is used to collect the sensor voltage, heating rod current, and pump battery internal resistance of the oxygen sensor under test. The phase space reconstruction module is used to calculate the second derivative of the sensor voltage. Based on the preset historical good product statistical parameters, the sensor voltage, heating rod current, pump battery internal resistance and the second derivative of the voltage are normalized to generate a dimensionless phase space state vector. The manifold analysis module is used to determine the actual evolution rate of the dimensionless phase space state vector; calculate the integral of the difference between the actual evolution rate and the ideal evolution rate defined by the preset good manifold vector field, and generate a cumulative trajectory deviation score. The feedback control module is used to compare the cumulative trajectory deviation score with the preset quality judgment threshold; in response to the cumulative trajectory deviation score being greater than the quality judgment threshold, the oxygen sensor to be tested is determined to be defective; in response to the cumulative trajectory deviation score being less than or equal to the quality judgment threshold, the oxygen sensor to be tested is determined to be qualified.

[0004] Preferably, the phase space reconstruction module performs normalization processing and is configured as follows: The original physical vector is constructed by calling the sensor voltage, heating rod current, pump battery internal resistance, and voltage second derivative; Access historical good product statistics parameters; Among them, the statistical parameters of historical good products include the mean vector and standard deviation diagonal matrix of the historical good product dataset; Based on the mean vector and the standard deviation diagonal matrix, the original physical vector is standardized to generate a dimensionless phase space state vector.

[0005] Preferably, the manifold analysis module generates a cumulative trajectory deviation score, including: The actual evolution rate is determined by differential calculation; Input the dimensionless phase space state vector into the good manifold vector field to obtain the ideal evolution rate under the current state; Calculate the Euclidean distance between the actual evolution rate and the ideal evolution rate in multidimensional space; Integrate the Euclidean distance over the preset test cutoff time to generate a cumulative trajectory deviation score.

[0006] Preferably, after determining that the oxygen sensor to be tested is defective, the feedback control module is further configured to: Calculate the average difference between the actual evolution rate and the ideal evolution rate within the test cutoff time, and generate an average deviation vector; The average deviation vector includes voltage, current, internal resistance, and acceleration components. Compare the magnitudes of the components in the average deviation vector to determine the direction of the largest component; In response to the fact that the direction of the maximum component is the voltage component, an adjustment command for the electrode coating process is generated; In response to the direction of the maximum component being the internal resistance component, an adjustment command for the sintering furnace temperature is generated. In response to the fact that the direction of the maximum component is the current component, an adjustment command is generated for the heating rod winding process or the lead soldering point. In response to the direction of the maximum component being the acceleration component, an adjustment command is generated for the porosity of the protective layer or the air pressure for spraying the breathable membrane, so as to regulate the diffusion rate of the gas.

[0007] Preferred options also include: The cloud-based model update module is used to respond to changes in raw material batches and obtain sample data collected in the new batch. Based on the sample data collected in the new batch, the vector field of the good product manifold is incrementally updated to generate the updated vector field model parameters.

[0008] Preferably, the cloud-based model update module performs incremental updates, including: Calculate the divergence between the sample data collected in the new batch and the data from the old batch; Adjust the learning rate for model updates based on the divergence value; By using the learning rate, the parameter matrix of the good product manifold vector field is updated by minimizing the trajectory prediction error of the sample data collected in the new batch.

[0009] Preferably, the data acquisition module is configured as follows: The oxygen sensor under test is simultaneously acquired through multiple channels using a high-speed acquisition card on the edge side at a preset high-frequency sampling frequency. The data acquisition time is limited to the transient phase before the oxygen sensor to be measured has established physical thermal equilibrium.

[0010] Compared with the prior art, the present invention has the following beneficial effects: 1. This system effectively solves the technical problem of slow production cycle caused by the long waiting time for thermal equilibrium in existing technologies by limiting data acquisition to the transient stage before the oxygen sensor has established physical thermal equilibrium and using a high-speed acquisition card on the edge side for multi-channel synchronous high-frequency sampling. By transforming the mode that relies on long-term steady-state judgment into a short-time high-frequency transient analysis mode, and using high-density data to construct smooth dynamic trajectories, the single-item testing cycle is significantly compressed, and the overall testing efficiency of the production line is greatly improved. 2. This system introduces phase space reconstruction and manifold analysis techniques to transform one-dimensional time series with different physical dimensions into dimensionless state points in a high-dimensional geometric space. It generates a cumulative trajectory deviation score by calculating the integral of the difference between the actual evolution rate and the ideal evolution rate defined by the good manifold vector field. This method overcomes the problem that the data in the transient start-up stage exhibits highly nonlinear and chaotic characteristics, making it difficult to make effective judgments using traditional linear thresholds. It can sensitively capture minute dynamic anomalies of the sensor during the heating process and achieve high-precision identification of hidden defects. 3. The feedback control module of this system has a reverse tracing function. After determining that a product is defective, it can accurately locate the specific process link that caused the quality abnormality by analyzing the values ​​of components such as voltage, current, internal resistance and acceleration in the average deviation vector. Based on this, the system automatically generates targeted equipment adjustment instructions, realizing automatic closed-loop control from quality inspection to production equipment parameter adjustment, effectively improving the stability of the production process. 4. This system achieves adaptive incremental updates for changes in raw material batches through a cloud-based model update module. It dynamically adjusts the model learning rate by calculating the divergence value of sample data from new and old batches, and uses new data to optimize the manifold vector field of good products online. This mechanism solves the problem that traditional algorithms require frequent manual adjustment of thresholds when facing fluctuations in material properties, ensuring that the system can automatically adapt to changes in the thermodynamic properties of raw materials and maintain a high level of prediction accuracy and generalization ability. Attached Figure Description

[0011] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0013] Example 1: Please see Figure 1 An automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform, including: The data acquisition module is used to collect the sensor voltage, heating rod current, and pump battery internal resistance of the oxygen sensor under test. The phase space reconstruction module is used to calculate the second derivative of the sensor voltage. Based on the preset historical good product statistical parameters, the sensor voltage, heating rod current, pump battery internal resistance and the second derivative of the voltage are normalized to generate a dimensionless phase space state vector. The manifold analysis module is used to determine the actual evolution rate of the dimensionless phase space state vector; calculate the integral of the difference between the actual evolution rate and the ideal evolution rate defined by the preset good manifold vector field, and generate a cumulative trajectory deviation score. The feedback control module is used to compare the cumulative trajectory deviation score with the preset quality judgment threshold; in response to the cumulative trajectory deviation score being greater than the quality judgment threshold, the oxygen sensor to be tested is determined to be defective; in response to the cumulative trajectory deviation score being less than or equal to the quality judgment threshold, the oxygen sensor to be tested is determined to be qualified.

[0014] The data acquisition module is configured as follows: The oxygen sensor under test is simultaneously acquired through multiple channels using a high-speed acquisition card on the edge side at a preset high-frequency sampling frequency. The data acquisition time is limited to the transient phase before the oxygen sensor to be measured has established physical thermal equilibrium.

[0015] This embodiment provides an automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform. The system aims to solve the technical problems in the prior art where oxygen sensor testing must wait for physical thermal equilibrium, resulting in a slow production cycle, and the problem that transient data chaos makes it difficult to pass the linear threshold for judgment. The system mainly includes a data acquisition module, a phase space reconstruction module, a manifold analysis module, and a feedback control module. As the source of data input, the data acquisition module is configured to use a high-speed acquisition card at the edge, such as a data acquisition card based on a field-programmable gate array (FPGA) architecture, to sample at a preset high-frequency frequency. Multi-channel synchronous data acquisition is performed on the oxygen sensor to be measured; in this embodiment, the sampling frequency is... The data acquisition frequency is set to 2kHz, and the acquisition time is strictly limited to the transient phase after the oxygen sensor to be measured starts heating, specifically the first 3 seconds after startup; in this embodiment, this time window... The specific value depends on the sensor's heat capacity and heating power, and its physical cutoff point is defined as the pump battery internal resistance. The resistance first drops to the operating temperature range. In this embodiment, the time corresponds to 3 seconds in the LSU4.9 sensor. The selection of this acquisition time window is based on the following: for a typical zirconia oxygen sensor, the 0 to 3 seconds after heating start-up constitute the electrochemical ignition stage; during this period, the ceramic body temperature rises sharply from room temperature to above 300°C, the oxygen ion mobility undergoes a step change, and the sensor is in a nonlinear thermodynamic non-equilibrium state; when time... Afterward, the sensor usually enters the steady-state stage of PID temperature control. At this time, the fluctuation of physical quantities is mainly dominated by external control logic, making it difficult to reflect the microscopic defects of the material itself. Therefore, locking for the first 3 seconds can capture the intrinsic dynamic differences caused by material and process defects to the greatest extent. During this period, the oxygen sensor has not yet established physical thermal equilibrium; the physical quantities collected include the sensor voltage. Heating rod current and the internal resistance of the pump battery These physical quantities can characterize the electrochemical activity establishment process and thermodynamic state changes of the oxygen sensor during the heating process. The specific implementation methods of the phase space reconstruction module, manifold analysis module, and feedback control module will be described in detail in subsequent embodiments. Through the above hardware configuration and acquisition strategy, this system transforms the problem that originally relied on long-term steady-state determination into a short-term, high-frequency transient analysis problem, and uses high-density data points to construct smooth dynamic trajectories, thereby significantly compressing the testing time of individual products.

[0016] Example 2: The phase space reconstruction module performs normalization processing and is configured as follows: The original physical vector is constructed by calling the sensor voltage, heating rod current, pump battery internal resistance, and voltage second derivative; Access historical good product statistics parameters; Among them, the statistical parameters of historical good products include the mean vector and standard deviation diagonal matrix of the historical good product dataset; Based on the mean vector and the standard deviation diagonal matrix, the original physical vector is standardized to generate a dimensionless phase space state vector.

[0017] This embodiment details the specific operating logic of the phase space reconstruction module; the core task of this module is to transform one-dimensional time series with different physical dimensions into dimensionless state points in a high-dimensional geometric space; The phase space reconstruction module calculates the second voltage derivative of the sensor voltage. An acceleration dimension is introduced to capture the drastic nature of voltage changes. It should be noted that the acceleration mentioned in this embodiment specifically refers to voltage change acceleration, that is, the second derivative of voltage change with time, which physically characterizes the rate of change of electrochemical reaction, rather than the gravitational acceleration in mechanical kinematics. This dimension is introduced because the defects in the porous protective layer of the oxygen sensor cause a small change in gas diffusion resistance. This change is not obvious in the first derivative of voltage, but will produce significant high-frequency oscillations in the curvature characteristics of the second derivative. To avoid amplification of power frequency interference and high-frequency noise during the differentiation process, this embodiment abandons the traditional direct difference method and instead uses a five-point cubic smoothed Savitzky-Golay filter for calculation; the specific calculation formula is configured as follows: ; in, The sampling interval is defined as the time interval. By combining the weighted information from two sampling points before and after the sampling, high-frequency noise spikes are effectively filtered out while preserving the voltage change trend. (Module call time) Sensor voltage Heating rod current Pump battery internal resistance and the second derivative of voltage Construct the original physical vector ; To eliminate the interference of different physical units such as voltage, current, and resistance on subsequent distance calculations, this module performs adaptive normalization on the original physical vector based on preset historical good product statistical parameters. These historical good product statistical parameters are derived from the statistical results of the historical good product dataset distributed from the cloud, specifically including the mean vector of the historical good product dataset. and the standard deviation diagonal matrix ; where the mean vector Includes average voltage Average current of heating rod Mean internal resistance and mean acceleration Standard deviation diagonal matrix The diagonal elements are the voltage standard deviations. Current standard deviation internal resistance standard deviation and acceleration standard deviation ; Based on the above parameters, the module generates a dimensionless phase space state vector through linear transformation. The calculation formula is as follows: ; in, This is the normalized four-dimensional state vector. This is the inverse of the standard deviation diagonal matrix; through this standardization transformation, the generated... The fact that it follows a standard normal distribution means that the data in each dimension have equal weight in the geometric space, which effectively avoids the problem that larger physical quantities mask the characteristics of smaller physical quantities, and provides a reliable mathematical basis for subsequent manifold analysis.

[0018] Example 3: The manifold analysis module generates a cumulative trajectory deviation score, including: The actual evolution rate is determined by differential calculation; Input the dimensionless phase space state vector into the good manifold vector field to obtain the ideal evolution rate under the current state; Calculate the Euclidean distance between the actual evolution rate and the ideal evolution rate in multidimensional space; Integrate the Euclidean distance over the preset test cutoff time to generate a cumulative trajectory deviation score.

[0019] This embodiment details the specific process by which the manifold analysis module generates the cumulative trajectory deviation score; this module utilizes the principles of differential geometry to quantify the difference between the test sample and the ideal good sample in terms of dynamic behavior; The manifold analysis module determines the actual evolution rate of the dimensionless phase space state vector through difference calculations. Specifically, utilizing the current moment state vector Compared with the previous sampling time state vector Perform a first-order difference approximation; the calculation formula is as follows: This ensures that the actual evolution rate has the same physical dimensions as the ideal evolution rate output by the good manifold vector field; At the same time, the module will transfer the current dimensionless phase space state vector Input the preset good manifold vector field The good quality manifold vector field is a mapping function that defines the theoretical evolution rate that an ideal good quality sensor should possess at any state point; in this embodiment, the good quality manifold vector field Specifically, it is constructed as a three-layer fully connected neural network to achieve a highly nonlinear mapping from the state space to the velocity vector space; the network includes: Input layer: Contains 4 neurons, corresponding to the dimensionless phase space state vector of the input. ; Hidden layer: Contains 16 neurons, using the hyperbolic tangent function. As an activation function, it captures the local bending features of the manifold; Output layer: Contains 4 neurons, corresponding to the ideal evolution rate of the output. The output layer does not use an activation function; Its forward propagation mathematical expression is: ; in, This is the weight matrix from the input layer to the hidden layer. This is the hidden layer bias vector; This is the weight matrix from the hidden layer to the output layer. For the output layer bias vector; these parameter sets This constitutes the kernel of the good quality manifold vector field; this parameter set is used before the system is put into online operation. It needs to be acquired through an offline pre-training phase; the specific process is as follows: select no less than 1000 oxygen sensors that have been confirmed as good products through bench calibration, collect their start-up transient data under a standard test environment, and construct an initial good product dataset; define the loss function. ,in, The total number of samples; here, the actual evolution rate of the training samples. The same first-order difference formula is used in the online testing steps of this embodiment. The calculations are performed to ensure that the model training objectives and the online inference logic are consistent in their physical definitions. Network parameter set Xavier uniform distribution is used for initialization to prevent gradient vanishing or exploding during deep network training; the Adam optimization algorithm is used to iteratively train the network until the loss function converges to the preset minimum value, thereby establishing the initial good manifold vector field; this process ensures that the vector field can accurately describe the evolution law of physical quantities under ideal conditions. The module uses this to obtain the ideal evolution rate in the current state. ,Right now ; Module calculates actual evolution speed With ideal evolution speed Euclidean distance in four-dimensional space, and for a preset test cutoff time. Integrate the distance within the range to generate a cumulative trajectory deviation score. The calculation formula is as follows: ; in, The cumulative trajectory deviation score is used to quantitatively characterize the degree of dynamic anomaly of the sensor during startup. To test the cutoff time, it is set to 3 seconds in this embodiment; This represents the Euclidean norm; through integration, this score can accumulate minute dynamic anomalies of the sensor throughout the transient startup process; if the sensor has microscopic defects, even if its value is within the acceptable range, its trend will produce an angle or modulus difference with the flow field, leading to... This method, based on the consistency of the velocity field, is highly sensitive for detecting latent defects.

[0020] Example 4: After determining that the oxygen sensor to be tested is defective, the feedback control module is also configured as follows: Calculate the average difference between the actual evolution rate and the ideal evolution rate within the test cutoff time, and generate an average deviation vector; The average deviation vector includes voltage, current, internal resistance, and acceleration components. Compare the magnitudes of the components in the average deviation vector to determine the direction of the largest component; In response to the fact that the direction of the maximum component is the voltage component, an adjustment command for the electrode coating process is generated; In response to the direction of the maximum component being the internal resistance component, an adjustment command for the sintering furnace temperature is generated. In response to the fact that the direction of the maximum component is the current component, an adjustment command is generated for the heating rod winding process or the lead soldering point. In response to the direction of the maximum component being the acceleration component, an adjustment command is generated for the porosity of the protective layer or the air pressure for spraying the breathable membrane, so as to regulate the diffusion rate of the gas.

[0021] This embodiment details the reverse tracing function of the feedback control module after determining that a product is defective; this function is used to identify the specific process steps that cause the quality abnormality. When the cumulative trajectory deviation score When the quality threshold is exceeded, the feedback control module determines that the oxygen sensor under test is defective; at this time, the module further calculates the test cutoff time. The average difference between the actual evolution rate and the ideal evolution rate is used to generate an average deviation vector. The calculation formula is as follows: ; The average deviation vector It contains four components, namely voltage components. Current components Internal resistance component and acceleration components The module determines the direction of the maximum component by comparing the absolute values ​​of these four components. The mapping relationship between the deviation component and the process steps established in this embodiment is based on the following physicochemical mechanism: Although there is a coupling effect between the various physical parameters of the oxygen sensor, during the transient start-up phase, specific process defects can cause the corresponding physical characteristics to exhibit significant anisotropy in phase space; therefore, the deviation vector... The maximum projection value in a certain dimension indicates that the corresponding dynamic mechanism plays a dominant role in the abnormal evolution. Although physical parameters such as voltage, current, and internal resistance are strongly coupled in steady state, experiments show that in the transient high-frequency response stage of 0-3 seconds, the projection amplitude of the dynamic anomaly caused by a single specific process defect in the corresponding dimension of phase space is significantly greater than its coupled projection in other dimensions, thus satisfying the engineering assumption of single fault dominance. Based on this dominant characteristic, the following mapping relationship is established: voltage component The main related electrode coating process: the establishment of voltage depends on the catalytic activity of the three-phase interface; if the electrode coating thickness is uneven or the effective area is insufficient, it will lead to a lag in the establishment of Nernst potential, which will be directly reflected in the numerical deviation in the voltage dimension. Internal resistance component The main related sintering process: The internal resistance of the pump battery is mainly determined by the grain boundary resistance of the zirconia ceramic body; the deviation of sintering temperature or holding time will change the ceramic grain size and density, thereby changing the migration resistance of oxygen ions between the lattice, which is directly reflected as the deviation of the internal resistance dimension. Current component The main related heating rod process: In constant voltage power supply mode, the heating rod current is directly affected by the cold resistance of the heating wire and the contact resistance of the lead wire. Therefore, this component is most sensitive to the winding pitch and solder joint quality. acceleration component The key related process for the permeable membrane is as follows: As mentioned earlier, the second derivative of the voltage is extremely sensitive to the turbulence effect during gas diffusion. Abnormal porosity of the protective layer or abnormal air pressure during permeable membrane spraying will alter the flow resistance characteristics of the exhaust gas reaching the electrode surface, leading to abnormal inflection points or fluctuations in the voltage response curve, thus generating a maximum projection on the acceleration component. The determination of the direction of this maximum projection is achieved using… Calculated, ignoring the influence of secondary coupling components; If the direction of the maximum component is the voltage component This typically indicates an abnormality in the sensor's Nernst potential buildup process, corresponding to issues with the effective electrode area or catalytic activity. Based on this, the module generates adjustment instructions for the electrode coating process, such as adjusting the spray thickness parameters; if the direction of the maximum component is the internal resistance component... This typically characterizes the abnormal ionic conductivity of zirconia ceramics, corresponding to ceramic sintering density issues. Based on this, the module generates adjustment instructions for the sintering furnace temperature or heating curve. By analyzing the directional characteristics of the deviation vector, this system achieves automatic closed-loop control from quality inspection to production equipment parameter adjustment.

[0022] Example 5: Also includes: The cloud-based model update module is used to respond to changes in raw material batches and obtain sample data collected in the new batch. Based on the sample data collected in the new batch, the vector field of the good product manifold is incrementally updated to generate the updated vector field model parameters.

[0023] The cloud-based model update module performs incremental updates, including: Calculate the divergence between the sample data collected in the new batch and the data from the old batch; Adjust the learning rate for model updates based on the divergence value; By using the learning rate, the parameter matrix of the good product manifold vector field is updated by minimizing the trajectory prediction error of the sample data collected in the new batch.

[0024] This embodiment details the incremental update strategy of the cloud model update module in response to changes in raw material batches; this module solves the problem that traditional algorithms require frequent manual adjustment of thresholds when facing material fluctuations; Based on the sample data collected in the new batch, the mean vector is recalculated and updated. and standard deviation diagonal matrix To ensure the dimensionless phase space state vector generated subsequently The new batch still follows the standard normal distribution characteristics; In response to a change in raw material batches, the cloud-based model update module acquires sample data from the new batch. To measure the differences in material properties between the old and new batches, the module calculates the divergence between the sample data from the new batch and the data from the old batch, for example, using KL divergence. Since the collected data are discrete time series points, divergence calculation cannot be performed directly; therefore, the module performs probability density estimation for the new batch sample set. and old batch sample set Using Gaussian kernel function respectively Construct the probability density function: ; in, The bandwidth parameter is set to 1.06 times the sample standard deviation; a probability density function is constructed for each dimension of the state vector to obtain the probability density function of the new and old batches in the first position. probability density function in each dimension and Then, the Jeffrey divergence for each dimension is calculated separately, and the mean of the four dimensions is taken as the final divergence value. : ; Based on this divergence value, the module dynamically adjusts the learning rate for model updates. When the divergence value is large, it indicates a significant change in material properties, and the system automatically increases the learning rate accordingly. To accelerate model adaptation; conversely, to reduce the learning rate to maintain model stability; specifically, the learning rate... with divergence value A nonlinear mapping relationship is established between them through the following S-shaped adjustment function: ; in, The preset base learning rate is 0.001. This is the maximum gain coefficient, with a value of 2.0, indicating a maximum allowable amplification of 3 times the learning rate; The divergence threshold is set to 0.5. To adjust the sensitivity, a value of 10 is used; this formula ensures that when... When the learning rate is much lower than the threshold, it remains constant. To maintain model stability, near; when When the learning rate significantly exceeds the threshold, it transitions quickly and smoothly to... This enables rapid capture of the characteristics of new batches; Using a defined learning rate The module updates the parameter matrix of the good product manifold vector field online by minimizing the trajectory prediction error of the newly collected sample data; the updated parameter matrix... The calculation is as follows: ; in, For the previous version Batch parameter matrix; For the updated number Batch parameter matrix; This represents the number of samples in the new batch. and The first in the new batch The state vector and actual velocity vector of each sample; This is the theoretical speed calculated based on current parameters; through the aforementioned incremental update mechanism, the system can automatically adapt to fluctuations in the thermodynamic properties of raw materials, ensuring consistently high-precision prediction capabilities and achieving adaptive evolution of the production model; the cloud-based model update module uses an encrypted communication protocol to transmit the updated parameter matrix. and the recalculated mean vector and standard deviation diagonal matrix The data is distributed to the edge acquisition card to complete the online replacement of the edge-side inference model.

[0025] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform, characterized in that, include: The data acquisition module is used to collect the sensor voltage, heating rod current, and pump battery internal resistance of the oxygen sensor under test. The phase space reconstruction module is used to calculate the second voltage derivative of the sensor voltage; Based on preset historical good product statistical parameters, the sensor voltage, heating rod current, pump battery internal resistance and voltage second derivative are normalized to generate a dimensionless phase space state vector. The manifold analysis module is used to determine the actual evolution rate of the dimensionless phase space state vector; Calculate the integral of the difference between the actual evolution velocity and the ideal evolution velocity defined by the preset good manifold vector field to generate a cumulative trajectory deviation score; The feedback control module is used to compare the cumulative trajectory deviation score with a preset quality judgment threshold. If the cumulative trajectory deviation score exceeds the quality judgment threshold, the oxygen sensor under test is determined to be defective. If the cumulative trajectory deviation score is less than or equal to the quality judgment threshold, the oxygen sensor under test is determined to be a qualified product.

2. The automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform as described in claim 1, characterized in that, The phase space reconstruction module performs normalization processing and is configured as follows: The original physical vector is constructed by calling the sensor voltage, heating rod current, pump battery internal resistance, and voltage second derivative; Access historical good product statistics parameters; Among them, the statistical parameters of historical good products include the mean vector and standard deviation diagonal matrix of the historical good product dataset; Based on the mean vector and the standard deviation diagonal matrix, the original physical vector is standardized to generate a dimensionless phase space state vector.

3. The automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform as described in claim 1, characterized in that, The manifold analysis module generates a cumulative trajectory deviation score, including: The actual evolution rate is determined by differential calculation; Input the dimensionless phase space state vector into the good manifold vector field to obtain the ideal evolution rate under the current state; Calculate the Euclidean distance between the actual evolution rate and the ideal evolution rate in multidimensional space; Integrate the Euclidean distance over the preset test cutoff time to generate a cumulative trajectory deviation score.

4. The automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform as described in claim 1, characterized in that, After determining that the oxygen sensor to be tested is defective, the feedback control module is further configured to: Calculate the average difference between the actual evolution rate and the ideal evolution rate within the test cutoff time, and generate an average deviation vector; The average deviation vector includes voltage, current, internal resistance, and acceleration components. Compare the magnitudes of the components in the average deviation vector to determine the direction of the largest component; In response to the fact that the direction of the maximum component is the voltage component, an adjustment command for the electrode coating process is generated; In response to the direction of the maximum component being the internal resistance component, an adjustment command for the sintering furnace temperature is generated. In response to the fact that the direction of the maximum component is the current component, an adjustment command is generated for the heating rod winding process or the lead soldering point. In response to the direction of the maximum component being the acceleration component, an adjustment command is generated for the porosity of the protective layer or the air pressure for spraying the breathable membrane, so as to regulate the diffusion rate of the gas.

5. The automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform as described in claim 1, characterized in that, Also includes: The cloud-based model update module is used to respond to changes in raw material batches and obtain sample data collected in the new batch. Based on the sample data collected in the new batch, the vector field of the good product manifold is incrementally updated to generate the updated vector field model parameters.

6. The automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform as described in claim 5, characterized in that, The cloud-based model update module performs incremental updates, including: Calculate the divergence between the sample data collected in the new batch and the data from the old batch; Adjust the learning rate for model updates based on the divergence value; By using the learning rate, the parameter matrix of the good product manifold vector field is updated by minimizing the trajectory prediction error of the sample data collected in the new batch.

7. The automated testing and inspection system for automotive oxygen sensor production lines based on a cloud platform as described in claim 1, characterized in that, The data acquisition module is configured as follows: The oxygen sensor under test is simultaneously acquired through multiple channels using a high-speed acquisition card on the edge side at a preset high-frequency sampling frequency. The data acquisition time is limited to the transient phase before the oxygen sensor to be measured has established physical thermal equilibrium.

Images