An edge-computing-based method for optimizing coating and drying process parameters of an automobile instrument panel

Abstract

The application discloses an automobile instrument panel coating drying process parameter optimization method based on edge computing. In the application, by actively injecting an adaptive perturbation signal into the chain speed, the traditional passive deviation adjustment is changed into active working condition detection, the local sensitivity coefficient obtained in real time enables the temperature power adjustment to compensate for the heat fluctuation caused by the chain speed variation in advance, thereby directly eliminating the inherent delay of the temperature loop after the residual error signal amplification, making the chain speed adjustment and temperature correction keep synchronization, significantly weakening the periodic residual rate deviation caused by the conveying speed fluctuation in the coating curing process, the response data calculated in real time is archived and used for model refitting, so that the control strategy is constantly close to the actual furnace type and coating characteristics with the increase of the running batch, thereby maintaining stable drying consistency under different seasons and different viscosity conditions.

Description

Technical Field

[0001] This invention belongs to the field of automotive instrument manufacturing technology, specifically a method for optimizing automotive instrument panel coating and drying process parameters based on edge computing. Background Technology

[0002] Automotive dashboard coating drying refers to a crucial process in automotive interior manufacturing where heat energy is used to physically dry and chemically cross-link the paint or coating applied to the dashboard surface. This process typically takes place in a drying tunnel or oven with precisely controlled temperature and air circulation. Through hot air convection, infrared radiation, or ultraviolet curing, the coating solvent evaporates and forms a dense protective film while ensuring the workpiece does not undergo thermal deformation. The drying process directly affects the dashboard's surface hardness, adhesion, weather resistance, and gloss, making it a decisive step in improving the appearance of interior parts, enhancing scratch resistance, and meeting vehicle VOCs environmental standards. It plays a core role in ensuring the quality of automotive interiors and the overall level of precision in vehicle manufacturing.

[0003] However, in existing technologies, drying processes generally use independent cascade control of chain speed and temperature. The temperature loop relies solely on residual rate deviation for hysteresis correction, and chain speed adjustment is often based on fixed beats or manual experience. There is a lack of real-time sensitivity and quantitative relationship between the two, making it impossible to predict the heat fluctuation range caused by chain speed changes in advance, resulting in frequent periodic over-baking or under-baking during the coating curing process. Summary of the Invention

[0004] The purpose of this invention is to provide a method for optimizing the coating drying process parameters of automotive dashboards based on edge computing in order to solve the problems mentioned above.

[0005] The technical solution adopted in this invention is as follows: A method for optimizing the coating drying process parameters of automotive dashboards based on edge computing, characterized by comprising the following steps:

[0006] S1: Temperature, wind speed and chain speed sensors are arranged on the inner wall of the drying oven and on both sides of the conveyor chain. The real-time signals are collected and transmitted to the buffer area of ​​the edge gateway.

[0007] S2: The edge gateway performs sliding window denoising and linear interpolation of missing values ​​on the original signal in the buffer area, and then aligns the multi-source data into a unified data frame sequence according to the spatial position of the workpiece in the furnace. Each frame contains the values ​​of multiple temperature measurement points under the current chain speed.

[0008] S3: Run a pre-trained lightweight drying quality prediction model at the edge. Taking the continuous frame sequence in step S2 as input, output the estimated values ​​of the current dashboard coating solvent residue rate and curing degree. S3 transmits the estimated values ​​to S5 as the control benchmark, and at the same time sends the estimated residual, i.e. the deviation from the measured value of the previous period, to S4 to trigger the dynamic adjustment of S4.

[0009] S4: Based on the fluctuation variance of S2 and the estimated residual of S3, an adaptive perturbation factor is constructed. A periodic perturbation signal is actively superimposed on the given chain speed. The change rate of the estimated residual fed back in step S3 is used to calculate the sensitivity of the perturbation to the coating residue rate in real time, thereby obtaining the local sensitivity coefficient between the current chain speed and temperature.

[0010] S5: Based on the estimated residual of S3 and the sensitivity coefficient of S4, an incremental PID algorithm is used to coordinately adjust the chain speed setpoint and the heating power of each zone. The chain speed adjustment adopts the suggested value of S4, and the temperature adjustment is obtained by proportional-integral transformation of the estimated residual, with amplitude limiting before output. S5 sends the control command to the field actuator and simultaneously sends the adjusted process parameter combination back to S2 so that S2 can mark the control state corresponding to this frame.

[0011] S6: After each batch of production is completed, the edge gateway packages the data sequences from all steps and uploads them to the cloud. The cloud uses these logs to periodically refit the prediction model and sends the updated model parameters to the edge to replace the old model in S3. After S6 completes the update, it notifies S1 to adjust the sampling priority and notifies S5 to clear the integral term to ensure that the control starting point of the next batch matches the model.

[0012] In a preferred embodiment, in step S1, eight K-type thermocouples are equidistantly installed along the conveying direction inside the drying oven, with each thermocouple head 150 mm vertically from the instrument panel surface. Simultaneously, four identical thermocouples are arranged on each side of the conveyor chain to capture cross-sectional temperature differences. A chain speed sensor is installed at the end of the drive roller, outputting 1024 pulse signals per revolution.

[0013] In a preferred embodiment, in step S2, the rising edge of the conveyor chain pulse is used as the frame synchronization signal. Each two adjacent pulses are defined as a spatial frame, corresponding to a fixed physical distance of 0.5 meters for the instrument panel movement. Temperature data falling within this frame is extracted from the original 100Hz sequence using linear interpolation. Wind speed data, due to its lower sampling rate of only 20Hz, is supplemented using the nearest neighbor hold method to reach the same number of points within the frame as the temperature data. After interpolation, each frame generates a 24-bit temperature value sequence and three wind speed value sequences, which, combined with the current average chain speed, form a complete frame object.

[0014] In a preferred embodiment, in step S3, the network input layer receives 12 consecutive frame objects from S2. Each frame object contains 24 temperature measurement points and 3 wind speed measurement points, therefore the input tensor dimension is 12 x 27. The network contains two one-dimensional convolutional layers: the first layer has a kernel size of 3 and 16 output channels, and the second layer has a kernel size of 5 and 8 output channels. A fully connected hidden layer with 4 neurons is then connected, and finally, the network splits into two independent output nodes, outputting the solvent residue rate estimate and the curing degree estimate, respectively.

[0015] In a preferred embodiment, in step S4, the fluctuation variance in the construction process of the adaptive disturbance factor reflects the turbulence intensity of the current hot air circulation in the furnace and the chattering degree of the chain speed itself, while the estimated residual directly indicates the deviation direction between the model output and the measured coating residual rate. When the fluctuation variance is large, it indicates that the system is being disturbed by external factors. At this time, the amplitude of the disturbance factor needs to be forcibly compressed to prevent the micro-disturbance signal from being submerged by noise and unable to extract an effective response. When the residual increases positively, that is, the actual residual rate is higher than the model prediction, the disturbance factor should be biased in the opposite direction to guide subsequent micro-disturbances to try to reduce the chain speed. In specific implementation, the fluctuation variance is first mapped to a confidence weight between zero and one, and then the estimated residual is converted into a signed driving intensity through nonlinear transformation. The product of the two is the adaptive disturbance factor for the current control cycle.

[0016] The formula for generating the adaptive perturbation factor is:

[0017]

[0018] in Let be the disturbance amplitude coefficient for the k-th control cycle. The preset maximum allowable amplitude scaling constant. The current estimated residual returned by S3. The residual gain coefficient is used to adjust the driving intensity. The variance of the signal fluctuation output by S2. The variance attenuation coefficient is used to quickly suppress the amplitude when the disturbance is strong.

[0019] In a preferred embodiment, in step S4, the adaptive disturbance factor is used as the amplitude coefficient of the sinusoidal perturbation signal and superimposed on the original setpoint of the chain speed. Simultaneously, it is ensured that the perturbation frequency is lower than the inherent cutoff frequency of the actuator, i.e., the variable frequency motor, to avoid triggering mechanical resonance. After the perturbation signal has been applied for two complete cycles, the system records the estimated residual value returned by S3 for each cycle during this period. Since fine-tuning the chain speed inevitably changes the residence time of the workpiece in the high-temperature zone, the coating residue rate will fluctuate accordingly. Based on this, the hysteresis phase and amplitude attenuation degree of the residue rate following the chain speed change can be quantitatively observed.

[0020] The formula for calculating the periodic perturbation signal superimposed on the given chain speed is:

[0021]

[0022] in The instantaneous offset from the chain speed setpoint. For the pre-selected perturbation angular frequency, The control sampling period for the edge gateway is specified here; the initial phase term is omitted to keep the expression concise.

[0023] In a preferred embodiment, in step S4, the first-order rate of change of the residual with respect to time is calculated using the residual sequence recorded during the perturbation period. This rate of change is divided by the perturbation amplitude of the current beat to obtain the sensitivity of the residual rate to chain speed perturbations. Considering that the furnace temperature is the fundamental source of drying heat, a change in chain speed is equivalent to changing the integral time for the workpiece to absorb heat. This integral effect is equivalent to an equivalent offset of the temperature setpoint. Therefore, the sensitivity is multiplied by the deviation term between the current furnace temperature and the setpoint, and historical corrections are accumulated to finally derive the equivalent temperature conversion factor for small changes in chain speed under the current operating conditions, which is the local sensitivity coefficient. This coefficient is directly output to S5 to guide the coordinated adjustment of heating power.

[0024] The formula for the first-order rate of change of the predicted residual is:

[0025]

[0026] in Let be the derivative of the residual with respect to time. This is the estimated residual value from the previous control cycle. This rate of change is used to measure the immediate response trend of coating residue under perturbation.

[0027] The formula for the sensitivity of coating residue rate to chain speed perturbations is:

[0028]

[0029] in For the current moment's response sensitivity, It is an extremely small positive number, which only plays a protective role when the absolute value of the denominator is below this threshold, preventing overflow when divided by zero.

[0030] The recursive formula for the local sensitivity coefficient between chain speed and temperature is:

[0031]

[0032] in This is the local sensitivity coefficient of the current output, representing the equivalent temperature value for each unit change in chain speed. Historical coefficients retained from the previous period. The recursive learning rate is used to control the update step size. This coefficient represents the real-time deviation between the current average furnace temperature and the process set temperature. It is updated every cycle and sent to S5. S5 then uses this coefficient to map the chain speed adjustment request into a compensation amount for heating power, thereby achieving coordinated control of air and heat.

[0033] In a preferred embodiment, in step S5, the chain speed fine-tuning amount transmitted from S4 is directly added to the chain speed setting value actually executed in the previous cycle to obtain the initial chain speed setting value for the current cycle. This initial setting value is then sent to the rate constraint device to limit its change per cycle to not exceed the maximum acceleration / deceleration threshold allowed by the mechanical transmission. The chain speed setting value after constraint is completed is immediately sent to the conveyor chain frequency converter, and the value is stored in the buffer as the accumulation reference for the next cycle. At this point, the chain speed adjustment path ends, and the system immediately switches to the temperature and power channel.

[0034] The recursive accumulation formula for the chain speed setpoint is:

[0035]

[0036] in This is the chain speed setting value issued for this test. The actual chain speed setting value for the previous beat. This is the fine-tuning amount of the chain speed directly input from S4.

[0037] In a preferred embodiment, in step S5, the temperature and power channel independently operates an incremental proportional-integral loop, which uses the estimated residual returned from S3 as the sole error driver. The proportional branch amplifies the difference between the current residual and the previous residual to capture the instantaneous trend of residual rate fluctuations; the integral branch directly amplifies the current residual to eliminate long-term accumulated static coating deviations. The sum of the two outputs yields the original value of the power adjustment increment. This original value is first subjected to a hard limiting circuit, forcing it to fall within the single-step change range of the burner valve's safe opening. After limiting, it is not directly issued but is additionally superimposed with a compensation increment, which is equal to the local sensitivity coefficient provided in S4 multiplied by the current chain speed fine-tuning amount and then multiplied by a negative coupling factor. This compensation forces the power adjustment direction to be opposite to the chain speed change direction to offset the drift in the total heat absorption of the workpiece caused by the chain speed change. Finally, the original increment after limiting and the compensation increment are added together to obtain the final adjustment percentage of temperature and power, which is then issued to the heating actuator.

[0038] The incremental proportional-integral formula for the temperature power regulation increment is:

[0039]

[0040] in This is the initial value of the power adjustment increment before limiting. This is the proportional gain coefficient. This is the integral gain coefficient. This is the estimated residual value for this auction. This is the estimated residual value of the previous shot.

[0041] The final output formula after amplitude limiting and wind temperature co-compensation is:

[0042]

[0043] in The final power adjustment percentage sent to the heating valve is represented by sat⋅, which indicates the amplitude saturation limiting calculation. λ is the hot air coupling suppression coefficient used to control the compensation intensity, and K_k is the local sensitivity coefficient input from S4, representing the equivalent temperature conversion value of the current chain speed variation. The chain speed fine-tuning amount is the same as that in Formula 1. This compensation term decouples the chain speed disturbance from the temperature channel in advance, so that the proportional-integral loop only needs to deal with the non-speed factors in the residual, thereby significantly reducing the risk of temperature overshoot and oscillation.

[0044] In a preferred embodiment, in step S6, the log file is retransmitted to the cloud object storage server using the local 5G module. After the transmission is completed, the cloud triggers a background task. This task merges the new log with the previous 50 batches of logs to form a training set. The Adam optimizer is used to refit the weights of the convolutional network used in S3. The learning rate is set to 0.001 and the number of iterations is 10. After training is completed, the cloud compresses the new weight parameters into a 32-megabyte model package and sends it to the edge gateway.

[0045] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0046] 1. In this invention, by actively injecting an adaptive perturbation signal into the chain speed, the traditional passive deviation adjustment is transformed into active working condition detection. The local sensitivity coefficient obtained in real time enables the temperature power adjustment to compensate for the heat fluctuation caused by chain speed changes in advance, thereby directly eliminating the inherent delay of the temperature loop waiting for the residual signal to be amplified before taking action. This keeps the chain speed adjustment and temperature correction synchronized, significantly reducing the periodic residual rate deviation caused by the fluctuation of the conveying speed during the coating curing process.

[0047] 2. In this invention, the amplitude of minor disturbances under external interference is dynamically suppressed using fluctuation variance, preventing excessive detection from affecting product quality when turbulence or chain speed vibration is severe. Simultaneously, sufficient excitation sensitivity is maintained under stable operating conditions for continuous optimization. Real-time calculated response data is archived and used for model refitting, ensuring that the control strategy continuously approximates the actual furnace type and coating characteristics as the number of operating batches increases, thereby maintaining stable drying consistency under different seasons and viscosity conditions. Attached Figure Description

[0048] Figure 1 This is a schematic diagram illustrating the process principle of the present invention;

[0049] Figure 2 This is a schematic diagram comparing the temperature control performance indicators in this invention;

[0050] Figure 3 This is a schematic diagram comparing the coating quality consistency indicators in this invention;

[0051] Figure 4 This is a schematic diagram comparing production response efficiency indicators in this invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] Example:

[0054] Reference Figure 1-4 A method for optimizing automotive dashboard coating drying process parameters based on edge computing includes the following steps:

[0055] S1: Temperature, wind speed and chain speed sensors are arranged on the inner wall of the drying oven and on both sides of the conveyor chain. The real-time signals are collected and transmitted to the buffer area of ​​the edge gateway.

[0056] S2: The edge gateway performs sliding window denoising and linear interpolation of missing values ​​on the original signal in the buffer area, and then aligns the multi-source data into a unified data frame sequence according to the spatial position of the workpiece in the furnace. Each frame contains the values ​​of multiple temperature measurement points under the current chain speed.

[0057] S3: Run a pre-trained lightweight drying quality prediction model at the edge. Taking the continuous frame sequence in step S2 as input, output the estimated values ​​of the current dashboard coating solvent residue rate and curing degree. S3 transmits the estimated values ​​to S5 as the control benchmark, and at the same time sends the estimated residual, i.e. the deviation from the measured value of the previous period, to S4 to trigger the dynamic adjustment of S4.

[0058] S4: Based on the fluctuation variance of S2 and the estimated residual of S3, an adaptive perturbation factor is constructed. A periodic perturbation signal is actively superimposed on the given chain speed. The change rate of the estimated residual fed back in step S3 is used to calculate the sensitivity of the perturbation to the coating residue rate in real time, thereby obtaining the local sensitivity coefficient between the current chain speed and temperature.

[0059] S5: Based on the estimated residual of S3 and the sensitivity coefficient of S4, an incremental PID algorithm is used to coordinately adjust the chain speed setpoint and the heating power of each zone. The chain speed adjustment adopts the suggested value of S4, and the temperature adjustment is obtained by proportional-integral transformation of the estimated residual, with amplitude limiting before output. S5 sends the control command to the field actuator and simultaneously sends the adjusted process parameter combination back to S2 so that S2 can mark the control state corresponding to this frame.

[0060] S6: After each batch of production is completed, the edge gateway packages the data sequences from all steps and uploads them to the cloud. The cloud uses these logs to periodically refit the prediction model and sends the updated model parameters to the edge to replace the old model in S3. After S6 completes the update, it notifies S1 to adjust the sampling priority and notifies S5 to clear the integral term to ensure that the control starting point of the next batch matches the model.

[0061] In step S1, eight K-type thermocouples are installed equidistantly along the conveying direction inside the drying oven. The vertical distance between the thermocouple head and the instrument panel surface is 150 mm. At the same time, four identical thermocouples are arranged on each side of the conveyor chain to capture the cross-sectional temperature difference. The chain speed sensor is installed at the end of the drive roller shaft and outputs 1024 pulse signals per revolution.

[0062] In step S2, the rising edge of the conveyor chain pulse is used as the frame synchronization signal. Each two adjacent pulses are defined as a spatial frame, corresponding to a fixed physical distance of 0.5 meters for the instrument panel movement. Temperature data falling within this frame is extracted from the original 100Hz sequence using linear interpolation. Wind speed data, due to its lower sampling rate of only 20Hz, is supplemented using the nearest neighbor hold method to reach the same number of points within the frame as the temperature data. After interpolation, each frame generates a 24-bit temperature value sequence and three wind speed value sequences, which, combined with the current average conveyor chain speed, form a complete frame object.

[0063] In step S3, the network input layer receives 12 consecutive frame objects from S2. Each frame object contains 24 temperature measurement points and 3 wind speed measurement points, so the input tensor dimension is 12 x 27. The network contains two one-dimensional convolutional layers: the first layer has a kernel size of 3 and 16 output channels, and the second layer has a kernel size of 5 and 8 output channels. A fully connected hidden layer with 4 neurons is then connected, and finally, the network splits into two independent output nodes, outputting the solvent residue rate estimate and the curing degree estimate, respectively.

[0064] In step S4, the fluctuation variance in the construction process of the adaptive disturbance factor reflects the turbulence intensity of the current hot air circulation in the furnace and the chattering degree of the chain speed itself, while the estimated residual directly indicates the deviation direction between the model output and the measured coating residual rate. When the fluctuation variance is large, it indicates that the system is under external interference. At this time, the amplitude of the disturbance factor needs to be forcibly compressed to prevent the micro-disturbance signal from being submerged by noise and unable to extract an effective response. When the residual increases positively, that is, the actual residual rate is higher than the model prediction, the disturbance factor should be biased in the opposite direction to guide subsequent micro-disturbances to try to reduce the chain speed. In specific implementation, the fluctuation variance is first mapped to a confidence weight between zero and one, and then the estimated residual is converted into a signed driving intensity through nonlinear transformation. The product of the two is the adaptive disturbance factor for the current control cycle.

[0065] The formula for generating the adaptive perturbation factor is:

[0066]

[0067] in Let be the disturbance amplitude coefficient for the k-th control cycle. The preset maximum allowable amplitude scaling constant. The current estimated residual returned by S3. The residual gain coefficient is used to adjust the driving intensity. The variance of the signal fluctuation output by S2. The variance attenuation coefficient is used to quickly suppress the amplitude when the disturbance is strong.

[0068] In step S4, the adaptive disturbance factor is used as the amplitude coefficient of the sinusoidal perturbation signal and superimposed on the original setpoint of the chain speed. Simultaneously, it is ensured that the perturbation frequency is lower than the inherent cutoff frequency of the actuator, i.e., the variable frequency motor, to avoid triggering mechanical resonance. After the perturbation signal has been applied for two complete cycles, the system records the estimated residual value returned by S3 for each cycle during this period. Since fine-tuning the chain speed inevitably changes the residence time of the workpiece in the high-temperature zone, the coating residue rate will fluctuate accordingly. Based on this, the hysteresis phase and amplitude attenuation of the residue rate following the chain speed change can be quantitatively observed.

[0069] The formula for calculating the periodic perturbation signal superimposed on the given chain speed is:

[0070]

[0071] in The instantaneous offset from the chain speed setpoint. For the pre-selected perturbation angular frequency, The control sampling period for the edge gateway is specified here; the initial phase term is omitted to keep the expression concise.

[0072] In step S4, the first-order rate of change of the residual with respect to time is calculated using the residual sequence recorded during the perturbation period. This rate of change is divided by the perturbation amplitude of the current beat to obtain the sensitivity of the residual rate to chain speed perturbations. Considering that the furnace temperature is the fundamental source of drying heat, a change in chain speed is equivalent to changing the integral time for the workpiece to absorb heat. This integral effect is equivalent to an equivalent offset of the temperature setpoint. Therefore, the response sensitivity is multiplied by the deviation term between the current furnace temperature and the setpoint and the historical correction is accumulated to finally derive the equivalent temperature conversion factor for small changes in chain speed under the current operating conditions, which is the local sensitivity coefficient. This coefficient is directly output to S5 to guide the coordinated adjustment of heating power.

[0073] The formula for the first-order rate of change of the predicted residual is:

[0074]

[0075] in Let be the derivative of the residual with respect to time. This is the estimated residual value from the previous control cycle. This rate of change is used to measure the immediate response trend of coating residue under perturbation.

[0076] The formula for the sensitivity of coating residue rate to chain speed perturbations is:

[0077]

[0078] in For the current moment's response sensitivity, It is an extremely small positive number, which only plays a protective role when the absolute value of the denominator is below this threshold, preventing overflow when divided by zero.

[0079] The recursive formula for the local sensitivity coefficient between chain speed and temperature is:

[0080]

[0081] in This is the local sensitivity coefficient of the current output, representing the equivalent temperature value for each unit change in chain speed. Historical coefficients retained from the previous period. The recursive learning rate is used to control the update step size. This coefficient represents the real-time deviation between the current average furnace temperature and the process set temperature. It is updated every cycle and sent to S5. S5 then uses this coefficient to map the chain speed adjustment request into a compensation amount for heating power, thereby achieving coordinated control of air and heat.

[0082] In step S5, the chain speed fine-tuning amount transmitted from S4 is directly added to the chain speed setting value actually executed in the previous cycle to obtain the initial chain speed setting value for this cycle. This initial setting value is then sent to the rate constraint device to limit its change per cycle to not exceed the maximum acceleration / deceleration threshold allowed by the mechanical transmission. After the constraint is completed, the chain speed setting value is immediately sent to the conveyor chain frequency converter, and at the same time, the value is stored in the buffer to be used as the accumulation reference for the next cycle. At this point, the chain speed adjustment path ends, and the system then switches to the temperature and power channel.

[0083] The recursive accumulation formula for the chain speed setpoint is:

[0084]

[0085] in This is the chain speed setting value issued for this test. The actual chain speed setting value for the previous beat. This is the fine-tuning amount of the chain speed directly input from S4.

[0086] In step S5, the temperature and power channel operates independently with an incremental proportional-integral (PI) loop, which uses the estimated residual returned from S3 as the sole error driver. The proportional branch amplifies the difference between the current residual and the previous residual to capture the instantaneous trend of residual rate fluctuations; the integral branch directly amplifies the current residual to eliminate long-term accumulated static coating deviations. The sum of the two outputs yields the original value of the power adjustment increment. This original value is first subjected to a hard limiting circuit, forcing it to fall within the single-step change range of the burner valve's safe opening. After limiting, it is not directly sent down, but an additional compensation increment is superimposed. This compensation increment is equal to the local sensitivity coefficient provided in S4 multiplied by the current chain speed fine-tuning amount and then multiplied by a negative coupling factor. This compensation forces the power adjustment direction to be opposite to the chain speed change direction to offset the drift in the total heat absorption of the workpiece caused by the chain speed change. Finally, the original increment after limiting and the compensation increment are added together to obtain the final temperature and power adjustment percentage, which is then sent to the heating actuator.

[0087] The incremental proportional-integral formula for the temperature power regulation increment is:

[0088]

[0089] in This is the initial value of the power adjustment increment before limiting. This is the proportional gain coefficient. This is the integral gain coefficient. This is the estimated residual value for this auction. This is the estimated residual value of the previous shot.

[0090] The final output formula after amplitude limiting and wind temperature co-compensation is:

[0091]

[0092] in The final power adjustment percentage sent to the heating valve is represented by sat⋅, which indicates the amplitude saturation limiting calculation. λ is the hot air coupling suppression coefficient used to control the compensation intensity, and K_k is the local sensitivity coefficient input from S4, representing the equivalent temperature conversion value of the current chain speed variation. The chain speed fine-tuning amount is the same as that in Formula 1. This compensation term decouples the chain speed disturbance from the temperature channel in advance, so that the proportional-integral loop only needs to deal with the non-speed factors in the residual, thereby significantly reducing the risk of temperature overshoot and oscillation.

[0093] In step S6, the log file is retransmitted to the cloud object storage server using the local 5G module. After the transmission is completed, a background task is triggered in the cloud. This task merges the new log with the previous 50 batches of logs to form a training set. The Adam optimizer is used to refit the weights of the convolutional network used in S3. The learning rate is set to 0.001 and the number of iterations is 10. After training is completed, the cloud compresses the new weight parameters into a 32-megabyte model package and sends it to the edge gateway.

[0094] Comparative example: The common furnace temperature cascade control and manual chain speed setting scheme used in the automotive painting industry is adopted.

[0095] Three K-type thermocouples are installed in the heating zone of the drying oven. The average readings of the three thermocouples are used as the main temperature feedback signal and sent to an independent PID controller. The controller output is connected to the proportional valve of the burner. The proportional gain Kp is set to 2.5, the integral time Ti is set to 40 seconds, the derivative time Td is set to 8 seconds, and the valve position output is limited to 20% to 80%. The conveyor chain speed is manually set by the workshop operator based on the viscosity cup measurement of the coating every morning. A viscosity value between 22 and 26 seconds corresponds to a chain speed of 0.8 meters per minute, and a viscosity value between 26 and 30 seconds corresponds to a chain speed of 0.6 meters per minute. Once set, it remains unchanged throughout the entire 8-hour shift. There is no electrical signal or data connection between the chain speed and temperature systems. The temperature PID is only responsible for maintaining a constant average furnace temperature and does not respond to the specific value of the chain speed. At the end of each batch, the quality inspector manually records the coating residue rate and the extreme value of the furnace temperature fluctuation, but this data is not fed back to the control system for adaptive adjustment in the next batch.

[0096] Performance comparison data and results analysis:

[0097] To demonstrate the actual improvement effect of this method, 12 continuously operating production batches were selected as the statistical objects, covering three common production disturbance scenarios: changes in coating viscosity, fluctuations in ambient temperature, and differences in the density of upstream incoming materials. Regarding temperature control performance, this method reduced the steady-state recovery time after disturbance from 22 minutes to 1.8 minutes, the average absolute deviation of furnace temperature from ±2.3 degrees Celsius to ±0.9 degrees Celsius, and the percentage of time with temperature fluctuations greater than ±3 degrees Celsius from 34% to 6%. Regarding coating quality consistency, the inter-batch range of residual rate was reduced from 6.3 percentage points to 2.1 percentage points, the intra-batch standard deviation from 1.7% to 0.6%, over-baking degradation was completely eliminated from 2.4 pieces per 100 pieces, and under-baking rework was reduced from 1.8 pieces per 100 pieces to 0.6 pieces. In terms of production response efficiency, the frequency of manual intervention by operators decreased from 4.2 times per shift to 0.3 times, the effective production time utilization rate increased from 86% to 97%, and the number of scraps during the batch-to-batch transition period decreased from 8.5 pieces to 1.2 pieces. These improvements stem from the fact that the S4 active perturbation detection in the method of this invention continuously provides the system with the sensitivity coefficient under the current operating conditions, enabling the S5 collaborative adjustment to compensate for heat drift caused by chain speed variations in advance, rather than relying on manual post-event intervention as in the comparative example.

[0098] Summary of the core differences between the comparative examples and the method of this invention:

[0099] The core flaw of the comparative method lies in the complete decoupling of temperature and chain speed. The temperature loop can only make lagging corrections based on furnace temperature deviations, and chain speed adjustment relies entirely on manual experience and cannot automatically respond to changes in operating conditions. Furthermore, it lacks active detection methods to obtain the real-time relationship between chain speed variations and coating residue rates. The method of this invention, however, actively acquires response sensitivity through periodic perturbation injection in S4, and inputs this information as a sensitivity coefficient into S5 to achieve air-heat coordination. Simultaneously, the data archiving and model refitting functions in S6 continuously improve the control strategy with each batch of operation. In the comparative method, manually recorded data is only used for quality inspection archiving and does not participate in the control loop closure. The method of this invention, however, uses each perturbation response as a valid sample for model updates, forming a closed-loop iteration of control and optimization.

[0100] from Figure 2 As can be seen, this method reduces the steady-state recovery time after disturbance from 22 minutes to 1.8 minutes. This is because the sensitivity coefficient calculated in real time by S4 enables S5 to superimpose the temperature compensation within the same step of chain speed variation, and the temperature loop can act in advance without waiting for the residual rate deviation to amplify. Figure 3 The data shows that the batch-to-batch range of residual rate was reduced from 6.3 percentage points to 2.1 percentage points, and the over-baking degradation was completely eliminated. This is directly due to the active perturbation detection continuously providing the system with the optimal operating point for the current operating conditions. Figure 4This further reflects a reduction in the frequency of manual intervention by operators from 4.2 times to 0.3 times, indicating that this method possesses adaptive adjustment capabilities in response to viscosity changes and fluctuations in coating density, no longer relying on manual experience for downtime adjustments. The three tables above, from the dimensions of temperature control, coating quality, and production efficiency, respectively confirm that the beneficial effects of this method in terms of active detection, coordinated adjustment, and adaptability are indeed present and significant compared to traditional fixed cascade control schemes.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0102] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing automotive dashboard coating drying process parameters based on edge computing, characterized in that: Includes the following steps: S1: Temperature, wind speed and chain speed sensors are arranged on the inner wall of the drying oven and on both sides of the conveyor chain. The real-time signals are collected and transmitted to the buffer area of ​​the edge gateway. S2: The edge gateway performs sliding window denoising and linear interpolation of missing values ​​on the original signal in the buffer area, and then aligns the multi-source data into a unified data frame sequence according to the spatial position of the workpiece in the furnace. Each frame contains the values ​​of multiple temperature measurement points under the current chain speed. S3: Run a pre-trained lightweight drying quality prediction model at the edge. Take the continuous frame sequence in step S2 as input and output the estimated values ​​of the current dashboard coating solvent residue rate and curing degree. S3 transmits the estimated values ​​to S5 as the control benchmark, and sends the estimated residual, i.e. the deviation from the measured value of the previous cycle, to S4 to trigger the dynamic adjustment of S4. S4: Based on the fluctuation variance of S2 and the estimated residual of S3, an adaptive perturbation factor is constructed. A periodic perturbation signal is actively superimposed on the given chain speed. The change rate of the estimated residual fed back in step S3 is used to calculate the sensitivity of the perturbation to the coating residue rate in real time, thereby obtaining the local sensitivity coefficient between the current chain speed and temperature. S5: Based on the estimated residual of S3 and the sensitivity coefficient of S4, an incremental PID algorithm is used to coordinate the adjustment of the chain speed setpoint and the heating power of each zone. The chain speed adjustment adopts the suggested value of S4, and the temperature adjustment is obtained by proportional integration from the estimated residual. The output is limited. S5 sends the control command to the field actuator and sends the adjusted process parameters back to S2 so that S2 can mark the control status corresponding to this frame. S6: After each batch of production is completed, the edge gateway packages the data sequences from all steps and uploads them to the cloud. The cloud uses these logs to periodically refit the prediction model and sends the updated model parameters to the edge to replace the old model in S3. After S6 is updated, it notifies S1 to adjust the sampling priority and notifies S5 to clear the integral term to ensure that the control starting point of the next batch matches the model.

2. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S1, eight K-type thermocouples are installed at equal intervals along the conveying direction inside the drying oven. The vertical distance between the head of each thermocouple and the surface of the instrument panel is 150 mm. At the same time, four thermocouples of the same type are arranged on each side of the conveying chain to capture the cross-sectional temperature difference. The chain speed sensor is installed at the end of the drive roller shaft and outputs 1024 pulse signals per revolution.

3. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S2, the rising edge of the conveyor chain pulse is used as the frame synchronization signal. Each two adjacent pulses are defined as a spatial frame, corresponding to a fixed physical distance of 0.5 meters for the instrument panel movement. Temperature data falling within the frame is extracted from the original 100 Hz sequence using linear interpolation. Wind speed data, due to its low sampling rate of only 20 Hz, is supplemented to the same number of points within the frame as the temperature data using the nearest neighbor hold method. After interpolation, each frame generates a temperature value sequence of length 24 and three wind speed value sequences, which, combined with the current average chain speed, form a complete frame object.

4. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S3, the network input layer receives 12 consecutive frame objects from S2. Each frame object contains 24 temperature measurement points and 3 wind speed measurement points, so the input tensor dimension is 12 times 27. The network contains two one-dimensional convolutional layers. The first convolutional kernel size is 3 and the number of output channels is 16. The second convolutional kernel size is 5 and the number of output channels is 8. Then, a fully connected hidden layer with 4 neurons is connected. Finally, it splits into two independent output nodes, which output the solvent residue rate estimate and the curing degree estimate, respectively.

5. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S4, the fluctuation variance in the construction process of the adaptive disturbance factor reflects the turbulence intensity of the current hot air circulation in the furnace and the chattering degree of the chain speed itself, while the estimated residual directly indicates the deviation direction between the model output and the measured coating residual rate. When the fluctuation variance is large, it indicates that the system is being disturbed by external factors. At this time, the amplitude of the disturbance factor needs to be forcibly compressed to prevent the micro-disturbance signal from being submerged by noise and unable to extract an effective response. When the residual increases positively, that is, the actual residual rate is higher than the model prediction, the disturbance factor should be biased in the opposite direction to guide the subsequent micro-disturbance to try to reduce the chain speed. In specific implementation, the fluctuation variance is first mapped to a confidence weight between zero and one, and then the estimated residual is converted into a signed driving intensity through nonlinear transformation. The two are multiplied to obtain the adaptive disturbance factor of the current control cycle. The formula for generating the adaptive perturbation factor is: in Let be the disturbance amplitude coefficient for the k-th control cycle. The preset maximum allowable amplitude scaling constant. The current estimated residual returned by S3. The residual gain coefficient is used to adjust the driving intensity. The variance of the signal fluctuation output by S2. The variance attenuation coefficient is used to quickly suppress the amplitude when the disturbance is strong.

6. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S4, the adaptive disturbance factor is used as the amplitude coefficient of the sinusoidal micro-disturbance signal and superimposed on the original given value of the chain speed. At the same time, it is ensured that the frequency of the micro-disturbance is lower than the inherent cutoff frequency of the actuator, i.e., the variable frequency motor, to avoid triggering mechanical resonance. After the micro-disturbance signal continues to act for two complete cycles, the system records the estimated residual value returned by S3 for each beat during this period. Since the chain speed fine adjustment will inevitably change the residence time of the workpiece in the high temperature zone, the coating residue rate will fluctuate at the same frequency. Based on this, the hysteresis phase and amplitude attenuation degree of the residue rate following the chain speed change can be quantitatively observed. The formula for calculating the periodic perturbation signal superimposed on the given chain speed is: in The instantaneous offset from the chain speed setpoint. For the pre-selected perturbation angular frequency, The control sampling period for the edge gateway is specified here; the initial phase term is omitted to keep the expression concise.

7. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S4, the first-order rate of change of the residual with respect to time is calculated using the residual sequence recorded during the perturbation period. This rate of change is divided by the perturbation amplitude of the current beat to obtain the sensitivity of the residual rate to chain speed perturbation. Considering that the furnace temperature is the fundamental source of drying heat, a change in chain speed is equivalent to changing the integral time for the workpiece to absorb heat. This integral effect is equivalent to an equivalent offset of the temperature setpoint. Therefore, the sensitivity of response is multiplied by the deviation term between the current furnace temperature and the setpoint and the historical correction is accumulated to finally derive the equivalent temperature conversion factor for a small change in chain speed under the current operating conditions, which is the local sensitivity coefficient. This coefficient is directly output to S5 to guide the coordinated adjustment of heating power. The formula for the first-order rate of change of the predicted residual is: in Let be the derivative of the residual with respect to time. This is the estimated residual value from the previous control cycle. This rate of change is used to measure the immediate response trend of the coating residue under perturbation. The formula for the sensitivity of coating residue rate to chain speed perturbations is: in For the current moment's response sensitivity, It is an extremely small positive number, which only plays a protective role when the absolute value of the denominator is lower than this threshold, preventing overflow when divided by zero; The recursive formula for the local sensitivity coefficient between chain speed and temperature is: in This is the local sensitivity coefficient of the current output, representing the equivalent temperature value for each unit change in chain speed. Historical coefficients retained from the previous period. The recursive learning rate is used to control the update step size. This coefficient represents the real-time deviation between the current average furnace temperature and the process set temperature. It is updated every time the process is completed and sent to S5. S5 then uses this coefficient to map the chain speed adjustment request into the compensation amount of heating power, thereby realizing the linkage control of air and heat.

8. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S5, the chain speed fine-tuning amount transmitted from S4 is directly added to the chain speed setting value actually executed in the previous cycle to obtain the initial chain speed setting value for this cycle. This initial setting value is then sent to the rate constraint device to limit its change in each cycle to not exceed the maximum acceleration or deceleration threshold allowed by the mechanical transmission. After the constraint is completed, the chain speed setting value is immediately sent to the conveyor chain frequency converter, and the value is stored in the cache to be used as the accumulation reference for the next cycle. At this point, the chain speed adjustment path ends, and the system immediately switches to the temperature and power channel. The recursive accumulation formula for the chain speed setpoint is: in This is the chain speed setting value issued for this test. The actual chain speed setting value for the previous beat. This is the fine-tuning amount of the chain speed directly input from S4.

9. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S5, the temperature power channel independently operates an incremental proportional-integral loop, which uses the estimated residual returned from S3 as the sole error driving source. The proportional branch amplifies the difference between the current residual and the previous residual to capture the instantaneous trend of residual rate fluctuations; the integral branch directly amplifies the current residual to eliminate long-term accumulated static coating deviations; the sum of the two outputs yields the original value of the power adjustment increment; this original value first passes through a hard limiting circuit to force it to fall within the single-step change range of the burner valve's safe opening; after limiting, it is not directly issued, but an additional compensation increment is superimposed, which is equal to the local sensitivity coefficient provided by S4 multiplied by the current chain speed fine-tuning amount and then multiplied by a negative coupling factor; The incremental proportional-integral formula for the temperature power regulation increment is: in This is the initial value of the power adjustment increment before limiting. This is the proportional gain coefficient. This is the integral gain coefficient. This is the estimated residual value for this auction. This is the estimated residual value from the previous shot; The final output formula after amplitude limiting and wind temperature co-compensation is: in The final power adjustment percentage sent to the heating valve is represented by sat⋅, which indicates the amplitude saturation limiting calculation. λ is the hot air coupling suppression coefficient used to control the compensation intensity, and K_k is the local sensitivity coefficient input from S4, representing the equivalent temperature conversion value of the current chain speed variation. The chain speed fine-tuning amount in the recursive cumulative formula of the chain speed setting is the same value.

10. The method for optimizing automotive dashboard coating drying process parameters based on edge computing according to claim 1, characterized in that: In step S6, the log file is retransmitted to the cloud object storage server using the local 5G module. After the transmission is completed, a background task is triggered in the cloud. This task merges the new log with the previous 50 batches of logs to form a training set. The Adam optimizer is used to refit the weights of the convolutional network used in S3. The learning rate is set to 0.001 and the number of iterations is 10. After training is completed, the cloud compresses the new weight parameters into a 32-megabyte model package and sends it to the edge gateway.

Images