Method suitable for environment bin heating test

Abstract

The invention provides a method suitable for an environmental bin heating test, and belongs to the technical field of environmental bin heating test, the method comprises the steps: determining a temperature measurement position through a graph theory topology optimization sensor arrangement algorithm, constructing a temperature field graph, configuring a Beckfu controller to collect temperature data at high frequency, and carrying out sliding window statistical test processing; a high-power main heater and a low-power auxiliary heater are installed at the top and the bottom of the environment bin respectively to form a closed-loop circulation system, temperature control partitions are divided according to the temperature deviation value, and a fuzzy sliding mode variable structure control algorithm is adopted to calculate partition heater control instructions. A control instruction is input into the thermal field dynamic balance optimization model for global collaborative optimization, an optimal power distribution scheme is output, a variable-power staged heating strategy is executed to achieve rapid temperature rise and fine temperature adjustment, and the technical problem that the test precision is reduced due to uneven temperature field distribution in the environment bin heating test process is solved.

Description

Technical Field

[0001] This invention belongs to the field of environmental chamber heating test technology, and more specifically, relates to a method suitable for environmental chamber heating tests. Background Technology

[0002] Environmental chamber heating tests are widely used in high-temperature performance testing in aerospace, electronic equipment, and materials science. Traditional heating control methods mainly employ fixed-power heaters combined with PID control algorithms to achieve temperature regulation. However, in existing technologies, due to factors such as irregular shapes of test specimens, differences in heat capacity, and varying heat dissipation rates in corner areas within the environmental chamber, a single control strategy struggles to simultaneously address the requirements of rapid heating and temperature uniformity, leading to localized overheating or underheating during the heating process. Traditional methods improve temperature uniformity by increasing the number of heaters or extending the heating time, but this increases energy consumption and reduces test efficiency. Furthermore, for complex operating conditions, there are still issues with inaccurate modeling of temperature field coupling relationships and a lack of global optimization in multi-heater collaborative control strategies. In other words, existing technologies suffer from the technical problem of reduced test accuracy due to uneven temperature field distribution during environmental chamber heating tests. Summary of the Invention

[0003] In view of this, the present invention provides a method for environmental chamber heating tests, which can solve the technical problem in the prior art where uneven temperature field distribution during environmental chamber heating tests leads to reduced test accuracy.

[0004] This invention is implemented as follows: A method suitable for environmental chamber heating tests is provided. A temperature field map topology sensor array is set up inside the environmental chamber. A graph theory topology optimization sensor placement algorithm is used to determine the installation positions of K-type thermocouple sensors. Real-time temperature data from all K-type thermocouple sensors in the temperature field map topology sensor array are collected. After performing sliding window statistical verification on the real-time temperature data to remove abnormal temperature values, the actual average temperature of the environmental chamber is calculated. A high-power main heater and a high-temperature resistant blower are installed at the top of the environmental chamber to form a closed-loop circulation system. Low-power auxiliary heaters are evenly arranged at the bottom of the environmental chamber. Based on the temperature deviation between the actual average temperature and the set temperature, the environmental chamber is divided into multiple temperature control zones. A fuzzy sliding mode variable structure control algorithm is used to calculate the zone heater control commands for each temperature control zone. The zone heater control commands for all temperature control zones are input into a dynamic equilibrium optimization model of the thermal field for global collaborative optimization, and the optimal power allocation scheme for each heater is output. Based on the optimal power allocation scheme, the conduction angle of the thyristor power regulator is adjusted to achieve precise power control of the high-power main heater and the low-power auxiliary heater, executing a variable power staged heating strategy.

[0005] Specifically, the graph theory topology optimization sensor placement algorithm involves dividing the environmental chamber into three-dimensional node grids by equally dividing the length, width, and height directions. It calculates the spatial distance between any two temperature nodes and the correlation coefficient of historical temperature data, using the historical temperature data correlation coefficient as the weight value of the edges between temperature nodes to construct a temperature field map. K-means clustering analysis is then performed on the temperature field map to divide it into multiple sub-regions with strong temperature correlation. Within each sub-region, a greedy algorithm is used to solve for the minimum dominance set, selecting the minimum representative node that can cover all temperature nodes in the sub-region as the installation location for the K-type thermocouple sensor.

[0006] The correlation coefficient of the historical temperature data is calculated using Pearson correlation analysis, and the expected number of sensors is preset based on the volume of the environmental chamber at a density of 0.5 to 1 sensor per cubic meter.

[0007] Specifically, the sliding window statistical test process involves establishing a sliding time window with a length of 30 sampling points, calculating the window mean and standard deviation for the real-time temperature data of each K-type thermocouple sensor within the sliding time window, performing a 3σ criterion test on the real-time temperature data of each K-type thermocouple sensor at the current sampling time to determine suspected abnormal temperature values, and further verifying the real-time temperature data of the K-type thermocouple sensors that are determined to be suspected abnormal temperature values ​​using the Grubbs test method.

[0008] The Grubbs statistic is calculated by dividing the absolute value of the difference between the suspected abnormal temperature value and the sample mean within the sliding time window by the sample standard deviation within the sliding time window. The Grubbs critical value is obtained by referring to the Grubbs critical value table based on the sliding time window length of 30 and the significance level of 0.05. The actual average temperature of the environmental chamber is obtained by removing the maximum and minimum temperature values ​​from all valid sensor temperature data after removing the abnormal temperature values ​​and then performing an arithmetic mean.

[0009] Specifically, the fuzzy sliding mode variable structure control algorithm defines the sliding mode switching function as a weighted sum of the temperature deviation value and the rate of change of the temperature deviation. When the absolute value of the sliding mode switching function is greater than the switching threshold, the sliding mode control law is used to calculate the control command for the partitioned heater. When the absolute value of the sliding mode switching function is less than or equal to the switching threshold, the algorithm switches to fuzzy control to calculate the control command for the partitioned heater.

[0010] In the sliding mode switching function, the weighting coefficient for the temperature deviation value is 7, the weighting coefficient for the rate of change of temperature deviation is 5, the switching threshold is 3℃, and the sliding mode gain coefficient is determined to be 15 based on the system uncertainty range and the upper limit of the disturbance.

[0011] Specifically, the fuzzy control establishes a fuzzy rule base for temperature deviation value and temperature deviation change rate. The fuzzy set of the fuzzy input variable for temperature deviation value is divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy set of the fuzzy input variable for temperature deviation change rate is also divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy set of the fuzzy output variable for power regulation is also divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large.

[0012] The input layer of the thermal field dynamic balance optimization model receives four types of feature data: temperature deviation value of each temperature control zone, temperature deviation change rate, current power ratio of each heater, and actual average temperature of the environmental chamber. The first hidden layer contains 128 neurons using the ReLU activation function, the second hidden layer contains 64 neurons combined with a multi-head self-attention mechanism, the third hidden layer contains 32 neurons, and the number of neurons in the output layer is equal to the total number of heaters, using the Softmax activation function.

[0013] The dynamic equilibrium optimization model for the thermal field employs a parameter regularization framework based on adaptive weight decay, calculating the gradient of parameters at each layer before each parameter update. Norm, will After normalizing the norm value to the interval of 0 to 1, multiply it by the basic attenuation coefficient to obtain the actual attenuation coefficient of the layer. In the parameter update formula, the parameter gradient is subtracted from the product of the parameter value and the actual attenuation coefficient.

[0014] The method for determining the number of attention heads in the multi-head self-attention mechanism is to divide the number of temperature control zones by 4 and round up to obtain the base number of heads, and then make corrections based on the temperature range in which the actual average temperature of the environmental chamber is located. When the actual average temperature of the environmental chamber is lower than the set temperature minus 50°C, the correction coefficient is 1.2; when the actual average temperature of the environmental chamber is between the set temperature minus 50°C and the set temperature minus 10°C, the correction coefficient is 1.0; and when the actual average temperature of the environmental chamber is higher than the set temperature minus 10°C, the correction coefficient is 0.8.

[0015] The training dataset establishment steps for the thermal field dynamic balance optimization model include placing test pieces of different shapes and sizes in the environmental chamber to conduct multiple sets of heating tests. During the heating test, the temperature deviation value, temperature deviation change rate, actual power ratio of each heater, and actual average temperature of the environmental chamber are recorded at a frequency of 10Hz as input features. At the same time, the power distribution scheme of each heater after manual adjustment by experienced operators and the corresponding temperature uniformity index are recorded as supervision labels.

[0016] The thermal field dynamic equilibrium optimization model is trained using the Adam optimizer for parameter updates, with an initial learning rate of 0.001. The loss function consists of two parts: mean squared error loss and regularization loss based on adaptive weight decay. In each training batch, forward propagation is used to calculate the output and loss value of the thermal field dynamic equilibrium optimization model, and back propagation is used to calculate the gradient of each layer's parameters. The layer weight decay coefficient is dynamically calculated based on the gradient norm of each layer to update the parameters of the thermal field dynamic equilibrium optimization model.

[0017] Specifically, the variable power phased heating strategy involves the following steps: in the initial stage of heating, all high-power main heaters and low-power auxiliary heaters operate simultaneously at their rated power to achieve rapid heating; when the actual average temperature of the environmental chamber reaches 85% of the set temperature, the middle stage begins, and the power ratio of each heater is adjusted in real time according to the temperature distribution of each temperature control zone; when the actual average temperature of the environmental chamber reaches 95% of the set temperature, the later stage begins, and the number of low-power auxiliary heaters operating is gradually reduced while the power of the high-power main heater is lowered to achieve fine temperature regulation.

[0018] Specifically, in the intermediate stage, the heater power of the temperature control zone with a temperature deviation value greater than 5°C is increased to 110% of the rated power; the heater power of the temperature control zone with a temperature deviation value between 2°C and 5°C is maintained at the rated power; and the heater power of the temperature control zone with a temperature deviation value less than 2°C is reduced to 70% of the rated power.

[0019] The conduction angle adjustment of the thyristor power regulator adopts an adaptive step size algorithm. The formula for calculating the conduction angle adjustment step value is the square root of the sum of the base step angle multiplied by the normalized temperature deviation value and the normalized temperature deviation change rate. The base step angle is set to 5 degrees. The normalized temperature deviation value is the quotient obtained by dividing the absolute value of the current temperature deviation value by the set temperature value. The normalized temperature deviation change rate is the quotient obtained by dividing the absolute value of the current temperature deviation change rate by 10℃ per second.

[0020] This invention employs a graph theory-based topology optimization sensor placement algorithm to construct a temperature field map and selects representative nodes as temperature measurement locations. A fuzzy sliding mode variable structure control algorithm enables rapid response and fine adjustment of zoned temperatures. A multi-head self-attention mechanism of a dynamic equilibrium optimization model of the thermal field is used to model the coupling relationships between each temperature control zone and output a globally coordinated power allocation scheme. Combined with a variable power staged heating strategy, the heater's operating state is dynamically adjusted at different stages of heating. The graph theory topology method ensures sufficient temperature field observation; fuzzy sliding mode control eliminates response lag under large deviations and control chattering under small deviations; the self-attention mechanism of the neural network model captures the spatial coupling characteristics of the temperature field; and the staged strategy balances heating rate and temperature uniformity. In summary, this invention solves the technical problem mentioned in the background art where uneven temperature field distribution leads to reduced experimental accuracy during environmental chamber heating experiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a schematic diagram of the heating system for the environmental chamber.

[0023] Figure 3 The graph shows the control process curves of the fuzzy sliding mode variable structure control algorithm.

[0024] Figure 4 The diagram shows the power distribution curve for the variable power staged heating strategy.

[0025] Figure 5 This is a graph showing the temperature uniformity during the heating test. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0027] like Figure 1 The diagram shown is a flowchart of a method for environmental chamber heating tests provided by the present invention. This method includes the following steps:

[0028] S01. Set up a temperature field map topology sensor array inside the environmental chamber, use graph theory topology optimization sensor layout algorithm to determine the installation position of K-type thermocouple sensor, discretize the environmental chamber space into temperature node grid and construct temperature field map, select representative nodes as the installation position of K-type thermocouple sensor through minimum dominant set algorithm and K-means clustering pre-partitioning.

[0029] S02. Configure Beckhoff CX2033 controller with EL3318 acquisition module. Beckhoff CX2033 controller acquires real-time temperature data of all K-type thermocouple sensors in temperature field topology sensor array at 1000Hz sampling rate. After performing sliding window statistical verification on the real-time temperature data to remove abnormal temperature values, calculate the actual average temperature of the environmental chamber.

[0030] S03. A high-power main heater and a high-temperature resistant blower are installed on the top of the environmental chamber to form a closed-loop circulation system. Small-power auxiliary heaters are evenly arranged at the bottom of the environmental chamber. The high-power main heater is equipped with a 30kW rated power thyristor power regulator, and the small-power auxiliary heater is equipped with a 1.5kW rated power thyristor power regulator. Heat circulation is achieved by using high-purity nitrogen as the heat transfer medium.

[0031] S04. Based on the temperature deviation between the actual average temperature and the set temperature of the environmental chamber, the environmental chamber is divided into multiple temperature control zones. Each temperature control zone is adjusted independently. The fuzzy sliding mode variable structure control algorithm is used to calculate the zone heater control command for each temperature control zone. When the absolute value of the temperature deviation is greater than the switching threshold, the sliding mode control law is used to calculate the zone heater control command. When the absolute value of the temperature deviation is less than or equal to the switching threshold, the fuzzy control is used to calculate the zone heater control command.

[0032] S05. Input the control commands of the zone heaters of all temperature control zones into the thermal field dynamic balance optimization model for global collaborative optimization. The thermal field dynamic balance optimization model outputs the optimal power allocation scheme for each heater. Beckhoff CX2033 controller achieves precise power control of the high-power main heater and low-power auxiliary heater by adjusting the conduction angle of the thyristor power regulator according to the optimal power allocation scheme.

[0033] S06. Implement a variable power phased heating strategy. In the initial stage of heating, all high-power main heaters and low-power auxiliary heaters work simultaneously at their rated power to achieve rapid heating. When the actual average temperature of the environmental chamber reaches 85% of the set temperature, the middle stage is entered. The power ratio of each heater is adjusted in real time according to the temperature distribution of each temperature control zone. When the actual average temperature of the environmental chamber reaches 95% of the set temperature, the later stage is entered. The number of low-power auxiliary heaters working is gradually reduced and the power of the high-power main heater is reduced to achieve fine temperature regulation.

[0034] The specific steps of the graph theory-based topology optimization sensor placement algorithm include: dividing the environmental chamber into three-dimensional node grids by equal division along the length, width, and height directions, with each intersection of the three-dimensional node grids serving as a temperature node; calculating the spatial distance between any two temperature nodes and the correlation coefficient of historical temperature data, using the historical temperature data correlation coefficient as the weight value of the edge between temperature nodes to construct a temperature field map; performing K-means clustering analysis on the temperature field map, with the number of clusters determined based on the volume of the environmental chamber and the expected number of sensors, dividing the temperature field map into multiple sub-regions with strong temperature correlation; using a greedy algorithm to solve for the minimum dominance set within each sub-region, selecting the fewest representative nodes that can cover all temperature nodes in the sub-region as the installation positions for K-type thermocouple sensors, ensuring that the selected set of K-type thermocouple sensor nodes has sufficient observation capability for the entire temperature field map. The historical temperature data correlation coefficient is obtained by Pearson correlation analysis of the temperature change curves at each temperature node location during historical heating experiments in the environmental chamber. The expected number of sensors is pre-set based on the volume of the environmental chamber at a density of 0.5 to 1 sensor per cubic meter.

[0035] The specific steps of the sliding window statistical test are as follows: A sliding time window with 30 sampling points is established. The window mean and standard deviation are calculated for the real-time temperature data of each K-type thermocouple sensor within the sliding time window. A 3σ criterion test is performed on the real-time temperature data of each K-type thermocouple sensor at the current sampling time. When the absolute value of the difference between the real-time temperature data of a K-type thermocouple sensor and the window mean of the K-type thermocouple sensor is greater than three times the window standard deviation, it is determined to be a suspected abnormal temperature value. The real-time temperature data of the K-type thermocouple sensors determined to be suspected abnormal temperature values ​​are further verified using the Grubbs test method. The Grubbs statistic of the suspected abnormal temperature value is calculated and compared with the Grubbs critical value. If the Grubbs statistic exceeds the Grubbs critical value, it is confirmed as an abnormal temperature value and removed. After removing the maximum and minimum temperature values ​​from all valid sensor temperature data after removing the abnormal temperature values, an arithmetic mean is calculated to obtain the actual average temperature of the environmental chamber. The calculation method for the Grubbs statistic is the absolute value of the difference between the suspected abnormal temperature value and the sample mean within the sliding time window divided by the sample standard deviation within the sliding time window. The Grubbs threshold is obtained by referring to the Grubbs threshold table based on a sliding time window length of 30 and a significance level of 0.05. The effective sensor temperature data consists of the real-time temperature data from the K-type thermocouple sensor after removing outlier temperature values.

[0036] The temperature deviation value is the difference between the actual average temperature of the environmental chamber and the set temperature. The switching threshold is 3℃.

[0037] The specific steps of the fuzzy sliding mode variable structure control algorithm include: defining the sliding mode switching function as a weighted sum of the temperature deviation value and the rate of change of the temperature deviation, with a weighting coefficient of 7 for the temperature deviation value and a weighting coefficient of 5 for the rate of change of the temperature deviation; when the absolute value of the sliding mode switching function is greater than the switching threshold, the sliding mode control law is used to calculate the control command for the partitioned heater, and the control output of the sliding mode control law is the product of the negative sliding mode gain coefficient and the sign function of the sliding mode switching function, with the sliding mode gain coefficient determined to be 15 based on the system uncertainty range and the upper bound of the disturbance; when the absolute value of the sliding mode switching function is less than or equal to the switching threshold, switching to fuzzy control to calculate the control command for the partitioned heater, and establishing the temperature deviation value and temperature... The system employs a fuzzy rule base for the deviation change rate. The fuzzy set of the fuzzy input variable for the temperature deviation value is divided into seven fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. Similarly, the fuzzy set of the fuzzy input variable for the temperature deviation change rate is also divided into seven fuzzy levels. The fuzzy output variable for the power regulation is also divided into seven fuzzy levels. Fuzzification is performed based on the current temperature deviation value and the temperature deviation change rate. Fuzzy inference is used to obtain the fuzzy output of the power regulation, and then the precise output value of the zone heater control command is obtained through centroid defuzzification. The temperature deviation change rate is calculated by dividing the difference in temperature deviation values ​​between two adjacent sampling times by the sampling time interval, which is 0.001 seconds. The system uncertainty range is determined based on the fluctuation range of the heat loss power in the environmental chamber, and the upper limit of the disturbance is determined based on the maximum amplitude of the change in the external ambient temperature of the environmental chamber.

[0038] The specific structure of the dynamic equilibrium optimization model for the thermal field is as follows: The input layer receives four types of feature data: temperature deviation value of each temperature control zone, temperature deviation change rate, current power ratio of each heater, and actual average temperature of the environmental chamber; the first hidden layer contains 128 neurons and uses the ReLU activation function to extract spatial features of temperature distribution; the second hidden layer contains 64 neurons and combines a multi-head self-attention mechanism to model the coupling relationship of the temperature fields of each temperature control zone. The number of attention heads in the multi-head self-attention mechanism is dynamically determined according to the number of temperature control zones and the actual average temperature of the environmental chamber; the third hidden layer contains 32 neurons for feature fusion; the number of neurons in the output layer is equal to the total number of heaters, and each output layer neuron corresponds to the target power ratio of a heater. The Softmax activation function is used to ensure that the sum of the power allocation of all heaters is 1. The dynamic equilibrium optimization model for the thermal field adopts a parameter regularization framework based on adaptive weight decay. An L2 regularization term is introduced into the loss function. The weight decay coefficient of each layer is not fixed but dynamically adjusted according to the gradient norm of the layer during backpropagation. Layers with larger gradient norms use larger weight decay coefficients to prevent overfitting, while layers with smaller gradient norms use smaller weight decay coefficients to retain learning ability. The specific implementation of the parameter regularization framework based on adaptive weight decay is as follows: before each parameter update, the L2 norm of the parameter gradient of each layer is calculated, the L2 norm value is normalized to the interval between 0 and 1 and then multiplied by the basic decay coefficient to obtain the actual decay coefficient of the layer, and the parameter gradient is subtracted from the product of the parameter value and the actual decay coefficient in the parameter update formula. The adaptive weight decay-based parameter regularization framework dynamically adjusts the regularization intensity of different layers, enabling the thermal field dynamic equilibrium optimization model to automatically balance the fitting and generalization abilities of each layer during training. This avoids insufficient feature extraction in shallow layers or overfitting in deep layers, significantly improving the adaptability of the thermal field dynamic equilibrium optimization model to different temperature field distributions and the accuracy of the optimal power allocation scheme. It ensures that the thermal field dynamic equilibrium optimization model can output reasonable heater collaborative control strategies even when facing irregular test specimen shapes, different heating rate requirements, and hot and cold spots at the edges of the environmental chamber. This effectively solves the deficiency of insufficient generalization ability of traditional fixed weight decay models under diverse operating conditions. Through a gradient-driven adaptive adjustment mechanism, it achieves refined overfit prevention and parameter optimization, enhancing model robustness while ensuring training convergence speed. This allows the optimal power allocation scheme to have stronger tolerance and more stable control performance in practical applications against sensor noise, operating condition deviations, and differences in test specimen heat capacity. The current power ratio of each heater is the ratio of the current actual power of each heater to its rated power.

[0039] The steps for establishing the training dataset for the thermal field dynamic equilibrium optimization model specifically include: placing test pieces of different shapes and sizes in the environmental chamber for multiple sets of heating tests, with different target temperatures and heating time requirements set for each set of heating tests; recording the temperature deviation value, temperature deviation change rate, actual power ratio of each heater, and actual average temperature of the environmental chamber at a frequency of 10Hz as input features during the heating tests; simultaneously recording the power distribution scheme of each heater after manual adjustment by experienced operators and the corresponding temperature uniformity index as supervision labels; cleaning the collected raw data to remove abnormal records during sensor failure periods; and dividing the cleaned data into training, validation, and test sets according to time series, with a ratio of 7:2:1. The temperature uniformity index is the standard deviation of the real-time temperature data of all K-type thermocouple sensors in the environmental chamber.

[0040] The specific steps for training the dynamic equilibrium optimization model of the thermal field include: updating parameters using the Adam optimizer with an initial learning rate of 0.001; the loss function consists of two parts: mean squared error loss and regularization loss based on adaptive weight decay. The mean squared error loss measures the difference between the output power allocation scheme of the dynamic equilibrium optimization model and the supervision label scheme, while the regularization loss based on adaptive weight decay prevents overfitting; in each training batch, forward propagation calculates the output and loss values ​​of the dynamic equilibrium optimization model of the thermal field, backpropagation calculates the gradient of each layer's parameters, dynamically calculates the layer weight decay coefficient based on the gradient norm of each layer, and updates the parameters of the dynamic equilibrium optimization model of the thermal field; every 10 epochs of training, the performance of the dynamic equilibrium optimization model of the thermal field is evaluated on the validation set. When the validation set loss does not decrease for 5 consecutive epochs, the learning rate is reduced by 0.5 times; training is terminated when the validation set loss does not decrease for 15 consecutive epochs; the parameters of the dynamic equilibrium optimization model of the thermal field with the smallest validation set loss are selected as the final model.

[0041] The method for determining the number of attention heads in the multi-head self-attention mechanism is as follows: Divide the number of temperature control zones by 4 and round up to obtain the base number of attention heads. Then, adjust the number of attention heads based on the actual average temperature range of the environmental chamber. The adjustment coefficient is 1.2 when the actual average temperature of the environmental chamber is lower than the set temperature minus 50℃; 1.0 when the actual average temperature of the environmental chamber is between the set temperature minus 50℃ and the set temperature minus 10℃; and 0.8 when the actual average temperature of the environmental chamber is higher than the set temperature minus 10℃. Multiply the base number of attention heads by the adjustment coefficient and round to obtain the actual number of attention heads. This dynamic adjustment mechanism enables the thermal field dynamic balance optimization model to increase the number of attention heads to capture more detailed temperature control zone coupling relationships when the temperature field distribution is significantly different in the early stages of heating, and to reduce the number of attention heads to decrease computational complexity and improve response speed when approaching the target temperature.

[0042] In the variable power phased heating strategy, the power of the main heater is 30kW during the initial heating phase, and the power of each auxiliary heater is 1.5kW. During the intermediate phase, the power is adjusted according to the temperature deviation of each temperature control zone. For temperature control zones with a deviation greater than 5℃, the heater power is increased to 110% of its rated power. For temperature control zones with a deviation between 2℃ and 5℃, the heater power remains at its rated power. For temperature control zones with a deviation less than 2℃, the heater power is reduced to 70% of its rated power. In the later phase, when the actual average temperature of the environmental chamber deviates from the set temperature by less than 5℃, one auxiliary heater is shut down every 30 seconds until only 1 to 2 auxiliary heaters remain operational. Simultaneously, the main heater power is reduced to 40% to 60% of its rated power for heat preservation control.

[0043] The conduction angle adjustment of the thyristor power regulator employs an adaptive step-size algorithm, determining the conduction angle adjustment step value based on the absolute value of the temperature deviation and the rate of change of the temperature deviation. The normalization method for the absolute value of the temperature deviation is to divide the current absolute value by the set temperature value; the normalization method for the rate of change of the temperature deviation is to divide the current absolute value of the rate of change of the temperature deviation by 10°C per second. The formula for calculating the conduction angle adjustment step value is the base step angle multiplied by the square root of the sum of the normalized temperature deviation value and the normalized rate of change of the temperature deviation, with the base step angle set to 5 degrees. When the absolute value of the temperature deviation is greater than 10°C or the absolute value of the rate of change of the temperature deviation is greater than 2°C per second, a large step is used for rapid power adjustment; when the absolute value of the temperature deviation is less than 3°C and the absolute value of the rate of change of the temperature deviation is less than 0.5°C per second, a small step is used for smooth transition to avoid temperature oscillations caused by power jumps. The normalized temperature deviation value is the quotient obtained by dividing the current absolute value of the temperature deviation by the set temperature value. The normalized temperature deviation change rate is the quotient obtained by dividing the absolute value of the current temperature deviation change rate by 10°C per second.

[0044] The closed-loop circulation system operates as follows: a high-temperature resistant blower blows high-purity air from the environmental chamber... After inhalation, the substance is heated to a high temperature by a high-power main heater, resulting in high-temperature, high-purity... High-temperature airflow is generated by injecting air into the environmental chamber through the outlet of a high-power main heater. As the high-temperature airflow flows within the chamber, it exchanges heat with the test specimen and the chamber walls, resulting in a temperature reduction. The cooled high-purity air... The heat is drawn back into the environment chamber by a high-temperature blower, creating a continuous circulation and ensuring the continuous transfer and even distribution of heat within the chamber. A low-power auxiliary heater at the bottom of the environment chamber compensates for heat loss in the bottom area, preventing excessive temperature differences between the top and bottom.

[0045] The specific implementation methods of the above steps are described in detail below.

[0046] The specific implementation of step S01 is as follows: First, the environmental chamber is divided into three equal parts along its length, width, and height, with each part set at a dividing point every 0.5 to 1 meter. This forms a three-dimensional node grid, and the intersection of each grid point constitutes a temperature node. Next, the Euclidean spatial distance between any two temperature nodes is calculated. Simultaneously, the temperature change curves of each temperature node are extracted from the historical heating test database of the environmental chamber. The Pearson correlation analysis method is used to calculate the correlation coefficient of historical temperature data between any two temperature nodes. This correlation coefficient is then used as the weight value of the edge between temperature nodes to construct a temperature field map. Then, K-means clustering analysis is performed on the temperature field map. The number of clusters is pre-set based on the volume of the environmental chamber, with a density of 0.5 to 1 sensor per cubic meter. The K-means clustering algorithm iteratively optimizes the temperature change pattern similarity of temperature nodes within the same cluster, dividing the temperature field map into multiple sub-regions with strong temperature correlation. In each sub-region, a greedy algorithm is used to find the minimum dominance set. The greedy algorithm selects the node that can cover the most uncovered temperature nodes in the sub-region as a candidate sensor location. This selection process is repeated until all temperature nodes in the sub-region are covered, thus ensuring that the selected set of K-type thermocouple sensor nodes has sufficient observation capability for the entire temperature field map. The graph theory topology optimization sensor placement algorithm aims to minimize the number of sensors while ensuring the accuracy of temperature field monitoring, avoiding the problem of insufficient sensors in areas with drastic temperature field changes and sensor redundancy in areas with gentle temperature field changes, which is common with traditional uniform placement methods.

[0047] The specific implementation of step S02 involves the Beckhoff CX2033 controller synchronously acquiring real-time temperature data from all K-type thermocouple sensors in the temperature field topology sensor array via an EL3318 acquisition module at a sampling rate of 1000Hz. The purpose of setting the sampling frequency to 1000Hz is to capture rapid changes in the temperature field. A sliding time window with a length of 30 sampling points is established. For each K-type thermocouple sensor's real-time temperature data within the sliding time window, the window mean and standard deviation are calculated. The window mean is obtained using the arithmetic mean method, and the window standard deviation is calculated using the sample standard deviation formula. The 3σ criterion is applied to the real-time temperature data of each K-type thermocouple sensor at the current sampling time. When the absolute value of the difference between a K-type thermocouple sensor's real-time temperature data and the window mean of that K-type thermocouple sensor is greater than three times the window standard deviation, it is determined to be a suspected abnormal temperature value. The 3σ criterion, based on the normal distribution theory, assumes that 99.7% of the data falls within the range of the mean plus or minus three standard deviations. Real-time temperature data from K-type thermocouple sensors identified as potentially abnormal temperature values ​​were further verified using the Grubbs' test. The absolute value of the difference between the suspected abnormal temperature value and the mean of the samples within the sliding time window was divided by the standard deviation of the samples within the sliding time window to obtain the Grubbs' statistic. Based on a sliding time window length of 30 and a significance level of 0.05, the Grubbs' critical value was found to be 2.745. If the Grubbs' statistic exceeded the critical value of 2.745, the temperature value was confirmed as abnormal and removed. After removing the maximum and minimum temperature values ​​from all valid sensor temperature data, the arithmetic mean was calculated to obtain the actual average temperature of the environmental chamber. The purpose of the sliding window statistical test is to identify and remove abnormal data caused by sensor malfunctions or localized sudden heat sources, thereby improving the robustness and accuracy of the calculation of the actual average temperature of the environmental chamber.

[0048] The specific implementation of step S03 involves installing a high-power main heater with a rated power of 30kW on the top of the environmental chamber. This high-power main heater uses resistance heating elements to convert electrical energy into heat energy. Simultaneously, a high-temperature resistant blower is installed at the outlet of the high-power main heater. The high-temperature resistant blower blows the high-purity air from the environmental chamber... After inhalation, the fluid is propelled through the heating zone of the high-power main heater, forming a closed-loop circulation system. Multiple low-power auxiliary heaters, each rated at 1.5kW, are evenly distributed at the bottom of the environmental chamber. The spacing between these low-power auxiliary heaters is based on the bottom area of ​​the environmental chamber and the number of auxiliary heaters. The high-power main heater is equipped with a 30kW rated power SCR power regulator, and the low-power auxiliary heaters are equipped with 1.5kW rated power SCR power regulators. The SCR power regulators continuously adjust the heater power by adjusting the conduction angle of the AC current. The purpose of this closed-loop circulation system is to achieve high purity... Forced convection enables rapid heat transfer and uniform distribution, avoiding temperature field stratification caused by natural convection. At the same time, a low-power auxiliary heater compensates for heat loss in the bottom area of ​​the environmental chamber.

[0049] The specific implementation of step S04 involves first calculating the temperature deviation value by measuring the difference between the actual average temperature of the environmental chamber and the set temperature. Based on the spatial distribution characteristics of the temperature deviation value, the environmental chamber is divided into multiple temperature control zones. The division method involves determining the zone boundaries according to the clustering results of the temperature deviation values, with areas having similar temperature deviation values ​​grouped into the same temperature control zone. The sliding mode switching function is defined as the sum of the temperature deviation value multiplied by a weighting coefficient of 7 and the temperature deviation change rate multiplied by a weighting coefficient of 5. The temperature deviation change rate is calculated by dividing the difference in temperature deviation values ​​between two adjacent sampling times by the sampling time interval of 0.001 seconds. When the absolute value of the sliding mode switching function is greater than the switching threshold of 3°C, a sliding mode control law is used to calculate the zone heater control command. The output of the sliding mode control law is the product of a negative sliding mode gain coefficient of 15 and the sign function of the sliding mode switching function. The sign function takes a value of 1 when the sliding mode switching function is greater than 0 and a value of -1 when the sliding mode switching function is less than 0. When the absolute value of the sliding mode switching function is less than or equal to the switching threshold of 3℃, fuzzy control is switched to. A fuzzy rule base is established for the temperature deviation value and the rate of change of temperature deviation. The fuzzy input variables of temperature deviation value and the rate of change of temperature deviation are divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy output variable of power regulation is also divided into 7 fuzzy levels. Fuzzy inference is performed by querying the fuzzy rule base based on the current temperature deviation value and the rate of change of temperature deviation. The centroid method is used to defuzzify the fuzzy output to obtain the accurate control command output value of the zone heater. The fuzzy sliding mode variable structure control algorithm combines the strong robustness of sliding mode control and the smooth adjustment characteristics of fuzzy control. When the temperature deviation is large, sliding mode control provides a fast response, and when approaching the set temperature, fuzzy control eliminates the chattering phenomenon of sliding mode control.

[0050] The specific implementation of step S05 involves inputting the control commands of the heaters in all temperature control zones, along with the temperature deviation value, temperature deviation change rate, current power ratio of each heater, and actual average temperature of the environmental chamber, into the dynamic equilibrium optimization model of the thermal field. The current power ratio of each heater is the ratio of the current actual power of each heater to its rated power. The input layer of the dynamic equilibrium optimization model of the thermal field receives four types of feature data. The 128 neurons in the first hidden layer extract the spatial gradient features and local hotspot features of the temperature distribution through the ReLU activation function. The 64 neurons in the second hidden layer combine a multi-head self-attention mechanism to calculate the temperature field coupling relationship between each temperature control zone. The multi-head self-attention mechanism captures the mutual influence between different temperature control zones through matrix multiplication of the key matrix and value matrix of the query matrix. The number of attention heads is obtained by dividing the number of temperature control zones by 4 and rounding up to obtain the basic number of heads. Then, the actual number of attention heads is obtained by multiplying the actual average temperature of the environmental chamber by a correction coefficient and rounding it to the nearest whole number. The correction coefficient is 1.2 in the early stage of heating, 1.0 in the middle stage, and 0.8 in the later stage. The 32 neurons in the third hidden layer fuse multi-scale features, and the number of neurons in the output layer equals the total number of heaters. The target power percentage for each heater is output through the Softmax activation function, ensuring that the total power distribution of all heaters equals 1. The Beckhoff CX2033 controller calculates the target power value for each heater based on the target power percentage and achieves precise power control of the high-power main heater and low-power auxiliary heater by adjusting the conduction angle of the thyristor power regulator. The purpose of the aforementioned dynamic thermal balance optimization model is to achieve coordinated optimization control of multiple heaters, avoiding local overheating or underheating problems caused by independent control of each temperature control zone.

[0051] The specific implementation of step S06 involves executing a variable power phased heating strategy based on the relationship between the actual average temperature of the environmental chamber and the set temperature. In the initial stage of heating, when the actual average temperature of the environmental chamber is below 85% of the set temperature, all high-power main heaters operate at their rated power of 30kW, and all low-power auxiliary heaters operate at their rated power of 1.5kW to achieve rapid heating. When the actual average temperature of the environmental chamber reaches 85% of the set temperature, the intermediate stage begins. Power is adjusted according to the temperature deviation value of each temperature control zone. For temperature control zones with a temperature deviation greater than 5℃, the heater power is increased to 110% of its rated power. For temperature control zones with a temperature deviation between 2℃ and 5℃, the heater power remains at its rated power. For temperature control zones with a temperature deviation less than 2℃, the heater power is reduced to 70% of its rated power. When the actual average temperature of the environmental chamber reaches 95% of the set temperature, the later stage begins. When the deviation between the actual average temperature of the environmental chamber and the set temperature is less than 5℃, one low-power auxiliary heater is shut down every 30 seconds until only 1 to 2 low-power auxiliary heaters remain in operation. Simultaneously, the power of the high-power main heater is reduced to 40% to 60% of its rated power for heat preservation control. The purpose of the variable power staged heating strategy is to prioritize the heating rate in the initial stage of heating, balance the heating rate and temperature uniformity in the middle stage, and prioritize temperature stability in the later stage to avoid temperature overshoot.

[0052] It should be noted that one of the key technical ideas of this invention is to achieve intelligent optimization of sensor placement by combining graph theory topology optimization sensor placement algorithm with temperature field spatial correlation and minimum dominance set theory. Traditional uniform placement methods do not consider the spatial distribution of the temperature field. In areas with large temperature gradients, insufficient sensor density leads to decreased monitoring accuracy, while in areas with gentle temperature fields, sensor redundancy results in wasted costs. This invention constructs a temperature field map and uses a clustering algorithm to identify sub-regions with strong temperature correlation. Within each sub-region, it solves for the minimum dominance set to optimize sensor placement, enabling a limited number of sensors to cover key temperature monitoring points. This reduces the number of sensors while improving the accuracy of temperature field reconstruction, providing accurate and reliable feedback signals for subsequent temperature control.

[0053] The second key technical concept of this invention is that the fuzzy sliding mode variable structure control algorithm organically integrates the strong robustness of sliding mode control with the smooth adjustment characteristics of fuzzy control. Traditional proportional-integral-derivative (PID) control lacks sufficient adjustment accuracy when facing environmental heat loss power fluctuations and external environmental temperature disturbances, easily generating large steady-state errors and overshoot. Although sliding mode control is highly robust to parameter changes and external disturbances, high-frequency chattering near the sliding surface affects control quality. This invention, by designing a switching function and switching threshold, uses a sliding mode control law to provide a fast response and suppress disturbances when the temperature deviation is large, and switches to fuzzy control to eliminate chattering when approaching the set temperature, achieving a balance between speed and smoothness, significantly reducing temperature steady-state error and fluctuation amplitude.

[0054] The third key technical idea of ​​this invention is that the dynamic equilibrium optimization model of the thermal field adopts a parameter regularization framework based on adaptive weight decay to achieve global collaborative optimization of multiple heaters. Traditional independent zone control methods do not consider the thermal coupling relationship between each temperature control zone; changes in the heater power of one zone will be affected by high-purity... Convection affects the temperature distribution of adjacent zones, causing local regulation to lead to global temperature field oscillations. This invention models the temperature field coupling relationship of each temperature control zone through a multi-head self-attention mechanism, and combines it with dynamically adjusted weight attenuation coefficients to prevent model overfitting. This enables the model to output a reasonable power distribution scheme when facing different test specimen shapes and operating conditions, avoiding local overheating or underheating and improving overall temperature uniformity.

[0055] The synergistic effect of the three key technological approaches mentioned above lies in forming a closed-loop control system from precise perception to intelligent decision-making and then to refined execution. The graph theory-based topology optimization sensor placement algorithm provides high-quality temperature field observation data, offering accurate input information for the fuzzy sliding mode variable structure control algorithm and the thermal field dynamic balance optimization model. The fuzzy sliding mode variable structure control algorithm achieves rapid response and smooth adjustment at each temperature control zone level, while the thermal field dynamic balance optimization model coordinates the power distribution of each heater at the global level to eliminate coupling interference. The complementary and synergistic relationship between the two ensures the speed, stability, and uniformity of temperature control. The organic combination of these three technologies overcomes the technical bottlenecks of traditional heating methods in large-size environmental chambers, namely slow heating speed, poor temperature uniformity, and low control accuracy, achieving a unified approach of rapid heating, precise temperature control, and uniform heating.

[0056] It should be noted that this invention also solves the following technical problem: traditional heating control systems are prone to misinterpretation of control commands and affect the accuracy of temperature regulation when faced with sensor noise interference and abnormal data. This invention establishes a sliding time window with 30 sampling points, performs a 3σ criterion check on each sensor data point to initially screen for suspected outliers, and then uses the Grubbs test for secondary verification. After removing confirmed abnormal temperature values ​​and extreme values, the actual average temperature of the environmental chamber is calculated by arithmetic averaging. This multi-layered statistical verification mechanism effectively filters instantaneous sensor noise and drift faults, ensuring the reliability of the temperature feedback signal and providing an accurate input basis for subsequent control algorithms, avoiding the risk of overall control failure due to single-point sensor failure.

[0057] Furthermore, this invention addresses the problem of insufficient generalization ability of neural network models under diverse heating conditions, leading to poor adaptability of power allocation schemes. This invention employs a parameter regularization framework based on adaptive weight decay. Before each parameter update, the L2 norm of the gradient of each layer's parameters is calculated, normalized, and multiplied by the base decay coefficient to obtain the actual decay coefficient for that layer. Layers with larger gradient norms use larger weight decays to prevent overfitting, while layers with smaller gradient norms use smaller weight decays to preserve learning ability. This gradient-driven adaptive adjustment mechanism enables the model to automatically balance the fitting and generalization abilities of each layer during training, avoiding insufficient shallow feature extraction or overfitting in deep layers. This ensures that the model can output reasonable heater collaborative control strategies even when facing complex conditions such as irregular test specimen shapes, different heating rate requirements, and hot and cold spots at the edges of environmental chambers.

[0058] Specifically, the principle of this invention is as follows: The fundamental reason why this invention can solve the problem of uneven temperature field distribution lies in the establishment of a complete closed-loop optimization system from temperature field sensing to multi-heater collaborative control. Graph theory topology algorithms ensure the coverage of the entire temperature field by sensor nodes through minimum dominance sets, providing accurate temperature distribution information for subsequent control. Fuzzy sliding mode variable structure control utilizes the strong robustness of sliding mode control to quickly approach the target temperature when there is a large temperature deviation, and switches to fuzzy control when there is a small deviation to avoid high-frequency chattering, thus solving the contradiction that a single control law cannot simultaneously consider response speed and stability. The multi-head self-attention mechanism of the thermal field dynamic balance optimization model can learn the thermal coupling relationship between different temperature control zones, and the regularization framework based on adaptive weight decay prevents the model from overfitting to the operating conditions, enabling the power allocation scheme to have generalization ability when facing different test pieces and heating requirements, thereby achieving dynamic balance and uniform distribution of the temperature field.

[0059] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0060] The specific implementation of step S01 is as follows: When setting up a temperature field mapping topology sensor array inside the environmental chamber, firstly, the length, width, and height directions of the environmental chamber are equally divided to form a three-dimensional node grid, with each intersection of the three-dimensional node grid serving as a temperature node. The spatial distance between any two temperature nodes and the correlation coefficient of historical temperature data are calculated, and this historical temperature data correlation coefficient is used as the weight value of the edge between temperature nodes to construct the temperature field map. The historical temperature data correlation coefficient is obtained by performing Pearson correlation analysis on the temperature change curves at each temperature node location during historical heating experiments in the environmental chamber. The correlation coefficient calculation formula is:

[0061] ;

[0062] In the formula, Temperature node With temperature node The correlation coefficient between historical temperature data, ranging from -1 to 1, is dimensionless; Temperature node At any moment The temperature value, in °C; Temperature node At any moment The temperature value, in °C; Temperature node The historical average temperature, in °C; Temperature node The historical average temperature, in °C; This represents the total number of historical sampling points, with an empirical value of 1000 to 5000. This is the number of the first temperature node; This is the number of the second temperature node; This refers to the sampling time sequence number. K-means cluster analysis was performed on the temperature field map. The number of clusters was determined based on the environmental chamber volume and the expected number of sensors. The expected number of sensors was determined based on the environmental chamber volume according to the following formula: The density of installing 0.5 to 1 sensors is preset, and the temperature field map is divided into multiple sub-regions with strong temperature correlation. Within each sub-region, a greedy algorithm is used to solve for the minimum dominance set, and the fewest representative nodes that can cover all temperature nodes in the sub-region are selected as the installation positions of the K-type thermocouple sensors, ensuring that the selected set of K-type thermocouple sensor nodes has sufficient observation capability for the entire temperature field map.

[0063] The specific implementation of step S02 is as follows: The Beckhoff CX2033 controller is equipped with an EL3318 acquisition module. The Beckhoff CX2033 controller acquires real-time temperature data from all K-type thermocouple sensors in the temperature field topology sensor array at a sampling rate of 1000Hz. The specific steps of the sliding window statistical test are as follows: The Beckhoff CX2033 controller establishes a sliding time window with a length of 30 sampling points. The window mean and standard deviation are calculated for the real-time temperature data of each K-type thermocouple sensor within the sliding time window. The formula for calculating the window mean is:

[0064] ;

[0065] In the formula, K-type thermocouple sensor The window mean, in °C; K-type thermocouple sensor Within the sliding time window Real-time temperature data for each sampling point, in °C; The part numbered for the K-type thermocouple sensor; This represents the sampling point number within the sliding time window, ranging from 1 to 30. The formula for calculating the window standard deviation is:

[0066] ;

[0067] In the formula, K-type thermocouple sensor The window standard deviation is given in °C. A 3σ criterion test is performed on the real-time temperature data of each K-type thermocouple sensor at the current sampling time. A suspected abnormal temperature value is determined when the absolute value of the difference between the real-time temperature data of a certain K-type thermocouple sensor and the window mean of that K-type thermocouple sensor is greater than three times the window standard deviation. The determination condition is as follows:

[0068] ;

[0069] In the formula, K-type thermocouple sensor The real-time temperature data at the current sampling moment is in °C. The real-time temperature data from the K-type thermocouple sensor, which is identified as a suspected abnormal temperature value, is further verified using the Grubbs' test. The formula for calculating the Grubbs' statistic is:

[0070] ;

[0071] In the formula, K-type thermocouple sensor The Grubbs statistic is dimensionless. If the Grubbs statistic exceeds the Grubbs critical value, it is considered an abnormal temperature value and is removed. The Grubbs critical value is... The values ​​are obtained from the Grubbs' critical value table based on a sliding time window length of 30 and a significance level of 0.05. A value of 2.745 (dimensionless) is typically used. The actual average temperature of the environmental chamber is calculated by removing the maximum and minimum temperature values ​​from all valid sensor temperature data after removing outlier values, and then performing an arithmetic mean. The calculation formula is as follows:

[0072] ;

[0073] In the formula, The actual average temperature of the environmental chamber is expressed in °C. This represents the total number of valid sensors remaining after removing abnormal temperature values. For effective sensors Real-time temperature data, in °C; The sensor number corresponding to the maximum temperature value; This is the sensor number corresponding to the minimum temperature value.

[0074] The specific implementation method of step S03 is the same as described above, and will not be repeated in detail here.

[0075] The specific implementation of step S04 is as follows: Based on the temperature deviation between the actual average temperature and the set temperature of the environmental chamber, the environmental chamber is divided into multiple temperature control zones, and each temperature control zone is adjusted independently. The formula for calculating the temperature deviation is:

[0076] ;

[0077] In the formula, This is the temperature deviation value, in °C. The set temperature, in °C, is preset according to experimental requirements. A fuzzy sliding mode variable structure control algorithm is used to calculate the control commands for the zone heaters in each temperature control zone. The sliding mode switching function is defined as the weighted sum of the temperature deviation value and the rate of change of the temperature deviation, calculated using the following formula:

[0078] ;

[0079] In the formula, This is a sliding mode switching function, dimensionless. This represents the rate of change of temperature deviation, expressed in °C / s. The reference temperature change rate is empirically set at 1℃ / s. The temperature deviation change rate is calculated by dividing the difference in temperature deviation between two adjacent sampling times by the sampling time interval. The calculation formula is as follows:

[0080] ;

[0081] In the formula, This is the temperature deviation value at the current sampling time, in °C. This represents the temperature deviation value at the previous sampling time, in °C. The sampling time interval is empirically set to 0.001 seconds. When the absolute value of the sliding mode switching function is greater than the switching threshold, the sliding mode control law is used to calculate the control command for the zone heater. The formula for calculating the switching threshold is:

[0082] ;

[0083] In the formula, The switching threshold is dimensionless. The formula for calculating the sliding mode control law is:

[0084] ;

[0085] In the formula, The control output of the sliding mode control law is dimensionless. The sliding mode gain coefficient has an empirical value of 15 and is dimensionless. The sign function of the sliding mode switching function, when When the value is 1, The value is -1 when The value is 0, dimensionless. When the absolute value of the sliding mode switching function is less than or equal to the switching threshold, the system switches to fuzzy control to calculate the zone heater control command. A fuzzy rule base is established for the temperature deviation value and the rate of change of temperature deviation. The fuzzy set of the fuzzy input variable for the temperature deviation value is divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy set of the fuzzy input variable for the rate of change of temperature deviation is also divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy set of the fuzzy output variable for the power regulation is also divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. Fuzzification is performed based on the current temperature deviation value and the rate of change of temperature deviation. The fuzzy output of the power regulation is obtained through fuzzy inference, and then the precise zone heater control command output value is obtained by defuzzification using the centroid method.

[0086] The specific implementation of step S05 is as follows: The control commands for the heaters in all temperature control zones are input into the dynamic equilibrium optimization model of the thermal field for global collaborative optimization. The specific structure of the dynamic equilibrium optimization model of the thermal field is as follows: The input layer receives four types of feature data: temperature deviation value of each temperature control zone, temperature deviation change rate, current power ratio of each heater, and actual average temperature of the environmental chamber. The formula for calculating the current power ratio of each heater is:

[0087] ;

[0088] In the formula, For heater The current power percentage is dimensionless; For heater The current actual power, in kW; For heater Rated power, in kW; The heaters are numbered. The first hidden layer contains 128 neurons and uses the ReLU activation function to extract spatial features of the temperature distribution. The second hidden layer contains 64 neurons and uses a multi-head self-attention mechanism to model the coupling relationship of the temperature fields in each temperature control zone. The number of attention heads in the multi-head self-attention mechanism is dynamically determined based on the number of temperature control zones and the actual average temperature of the environmental chamber. The formula for calculating the basic number of heads is:

[0089] ;

[0090] In the formula, The number of basic heads is dimensionless. The number of temperature control zones, dimensionless; This is a rounding function. It is corrected based on the actual average temperature range of the environmental chamber, with a correction factor... The rule for determining the value is: when hour ,when hour ,when hour The correction factor is dimensionless. The formula for calculating the actual number of attention heads is:

[0091] ;

[0092] In the formula, The actual number of attention heads, dimensionless; This is a rounding function. The third hidden layer contains 32 neurons for feature fusion. The number of neurons in the output layer equals the total number of heaters, and each output layer neuron corresponds to the target power percentage of one heater. A Softmax activation function is used to ensure that the sum of the power distributions of all heaters is 1. The dynamic equilibrium optimization model of the thermal field adopts a parameter regularization framework based on adaptive weight decay. Before each parameter update, the L2 norm of the gradient of each layer's parameters is calculated, and the normalized calculation formula is:

[0093] ;

[0094] In the formula, For the first The normalized L2 norm of the gradient of the layer parameters is dimensionless. For the first The L2 norm of the gradient of the layer parameters is dimensionless. The maximum value of the L2 norm of the parameter gradients in all layers is dimensionless. Number the layers of the neural network; This is the index variable for traversing all layers. The actual attenuation coefficient is calculated using the following formula:

[0095] ;

[0096] In the formula, For the first The actual attenuation coefficient of the layer is dimensionless. The basic attenuation coefficient has an empirical value of 0.0001 and is dimensionless.

[0097] The specific implementation of step S06 is as follows: A variable power, phased heating strategy is executed. In the initial stage of heating, all high-power main heaters and low-power auxiliary heaters operate simultaneously at their rated power to achieve rapid heating. The power of the high-power main heater is 30kW, and the power of each low-power auxiliary heater is 1.5kW. When the actual average temperature of the environmental chamber reaches 85% of the set temperature, it enters the intermediate stage. The determination condition is:

[0098] ;

[0099] In the intermediate stage, the power ratio of each heater is adjusted in real time according to the temperature distribution of each temperature control zone. For temperature control zones with a temperature deviation greater than 5℃, the heater power is increased to 110% of the rated power. For temperature control zones with a temperature deviation between 2℃ and 5℃, the heater power remains at the rated power. For temperature control zones with a temperature deviation less than 2℃, the heater power is reduced to 70% of the rated power. The later stage begins when the actual average temperature of the environmental chamber reaches 95% of the set temperature, determined by the following criteria:

[0100] ;

[0101] In the later stages, the number of low-power auxiliary heaters is gradually reduced, and the power of the high-power main heater is lowered to achieve precise temperature control. When the absolute value of the deviation between the actual average temperature of the environmental chamber and the set temperature is less than 5°C, one low-power auxiliary heater is shut down every 30 seconds until only 1 to 2 low-power auxiliary heaters remain in operation. Simultaneously, the power of the high-power main heater is reduced to 40% to 60% of its rated power for heat preservation control. The Beckhoff CX2033 controller, based on the optimal power allocation scheme, achieves precise power control of the high-power main heater and low-power auxiliary heaters through the conduction angle adjustment of the thyristor power regulator. An adaptive step-size algorithm is used to determine the conduction angle adjustment step value. The normalized temperature deviation value is calculated using the following formula:

[0102] ;

[0103] In the formula, This is the normalized temperature deviation value, dimensionless. The formula for calculating the normalized temperature deviation rate of change is:

[0104] ;

[0105] In the formula, This is the normalized rate of change of temperature deviation, dimensionless. The temperature deviation change rate is a reference value, empirically set at 10℃ / s. The formula for calculating the conduction angle adjustment step value is:

[0106] ;

[0107] In the formula, The step value for adjusting the conduction angle is in degrees; The basic step angle is 5 degrees, based on experience. When the absolute value of the temperature deviation is greater than 10℃ or the absolute value of the temperature deviation change rate is greater than 2℃ / s, a large step is used to quickly adjust the power. When the absolute value of the temperature deviation is less than 3℃ and the absolute value of the temperature deviation change rate is less than 0.5℃ / s, a small step is used for smooth transition to avoid temperature oscillation caused by power step.

[0108] To better understand and implement this invention, a specific application scenario of the invention is provided below as an example 2: To verify the effectiveness of the invention, technicians built a test environment and conducted a high-temperature heating test on a certain type of avionics equipment. This test was conducted in a volume of... The experiment was conducted in an environmental chamber, which was 7m long, 5.2m wide, and 2.9m high. The goal was to raise the temperature in the environmental chamber from 25°C to 185°C while maintaining temperature uniformity. Technicians first installed a temperature field mapping topology sensor array inside the environmental chamber according to the method described in this invention. The chamber space was divided into 14 equal segments along its length, 10 equal segments along its width, and 6 equal segments along its height, forming 840 three-dimensional node grid intersections as temperature nodes. By analyzing historical heating test data, the Pearson correlation coefficient between any two temperature nodes was calculated as the edge weight to construct the temperature field map. Based on the volume of the environmental chamber, a density of 0.8 sensors per cubic meter was installed, with an expected number of sensors set to 86. Technicians performed K-means cluster analysis on the temperature field map, setting the number of clusters to 12, dividing the temperature field map into 12 sub-regions with strong temperature correlation. A greedy algorithm was used to solve for the minimum dominance set in each sub-region, and the installation positions of 83 K-type thermocouple sensors were finally determined. This set of sensor nodes has sufficient observation capability for the entire temperature field map.

[0109] Technicians configured a Beckhoff CX2033 controller with an EL3318 data acquisition module to collect real-time temperature data from all 83 K-type thermocouple sensors at a sampling rate of 1000Hz. Four high-power main heaters (30kW rated) and two high-temperature blowers were installed at the top of the environmental chamber to form a closed-loop circulation system. Eight low-power auxiliary heaters (1.5kW rated) were evenly distributed at the bottom of the environmental chamber. Figure 2 As shown, the high-power main heater is equipped with a 30kW rated power thyristor power regulator, and the low-power auxiliary heater is equipped with a 1.5kW rated power thyristor power regulator, passing through high-purity... The thermocouple sensor serves as a heat transfer medium for heat circulation. At the start of the experiment, the Beckhoff CX2033 controller established a sliding time window with 30 sampling points. The window mean and standard deviation were calculated for the real-time temperature data of each K-type thermocouple sensor. When the absolute value of the difference between a sensor's real-time temperature data and the window mean exceeded three times the window standard deviation, it was considered a suspected abnormal temperature value, which was further verified using the Grubbs test. The Grubbs statistic is calculated as the absolute value of the difference between the suspected abnormal temperature value and the sample mean within the sliding time window, divided by the sample standard deviation. Based on the window length of 30 and a significance level of 0.05, the Grubbs critical value was found to be 2.745. If the Grubbs statistic exceeded 2.745, it was confirmed as an abnormal temperature value and removed. After removing the maximum and minimum temperature values ​​from all valid sensor temperature data (after removing the abnormal values), an arithmetic mean was calculated to obtain the actual average temperature of the environmental chamber.

[0110] During the heating process, technicians divided the environmental chamber into six temperature control zones based on the temperature deviation between the actual average temperature and the set temperature of 185℃. Each temperature control zone was adjusted independently. For example... Figure 3 As shown, a fuzzy sliding mode variable structure control algorithm is used to calculate the control commands for the zone heaters in each temperature control zone. The sliding mode switching function is defined as the weighted sum of the temperature deviation value and the rate of change of the temperature deviation. The weighting coefficient for the temperature deviation value is 7, and the weighting coefficient for the rate of change of the temperature deviation is 5. The rate of change of the temperature deviation is calculated by dividing the difference in temperature deviation values ​​at two adjacent sampling times by the sampling time interval of 0.001 seconds. When the absolute value of the sliding mode switching function is greater than 3℃, the sliding mode control law is used to calculate the zone heater control commands, and the sliding mode gain coefficient is set to 15. When the absolute value of the sliding mode switching function is less than or equal to 3℃, fuzzy control is switched to. A fuzzy rule base for the temperature deviation value and the rate of change of the temperature deviation is established. The fuzzy sets of the fuzzy input variables and output variables are divided into seven fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy output of the power adjustment is obtained through fuzzy inference, and the accurate zone heater control command output value is obtained by defuzzification using the centroid method.

[0111] Technicians input the control commands for the heaters in all six temperature control zones into the dynamic equilibrium optimization model for global collaborative optimization. The input layer of the dynamic equilibrium optimization model receives four types of feature data: temperature deviation value, temperature deviation change rate, current power ratio of each heater, and actual average temperature of the environmental chamber. The first hidden layer contains 128 neurons that use the ReLU activation function to extract spatial features of the temperature distribution. The second hidden layer contains 64 neurons that combine a multi-head self-attention mechanism to model the coupling relationship of the temperature fields in each temperature control zone. The number of attention heads in the multi-head self-attention mechanism is dynamically determined based on the number of temperature control zones and the actual average temperature of the environmental chamber. The number of temperature control zones (6) is divided by 4 and rounded up to obtain a base number of 2. In the initial stage of heating, the actual average temperature of the environmental chamber is 42℃, which is lower than the set temperature minus 50℃ (135℃). The correction coefficient is 1.2, and the actual number of attention heads is 2 multiplied by 1.2 and rounded to 2. The third hidden layer contains 32 neurons for feature fusion. The number of neurons in the output layer is 12, equal to the total number of heaters. Each output layer neuron corresponds to the target power percentage of one heater, and a Softmax activation function is used to ensure that the sum of the power distributions of all heaters is 1. The dynamic equilibrium optimization model of the thermal field adopts a parameter regularization framework based on adaptive weight decay. Before each parameter update, the L2 norm of the gradient of each layer's parameters is calculated. After normalizing the L2 norm value to the interval between 0 and 1, it is multiplied by the base decay coefficient of 0.0001 to obtain the actual decay coefficient of that layer. In the parameter update formula, the parameter gradient is subtracted from the product of the parameter value and the actual decay coefficient, so that the model automatically balances the fitting ability and generalization ability of each layer during training.

[0112] like Figure 4As shown, technicians implemented a variable-power, phased heating strategy. In the initial heating phase, all four high-power main heaters and eight low-power auxiliary heaters operated simultaneously at their rated power to achieve rapid heating. The total power of the high-power main heaters was 120kW, and the total power of the low-power auxiliary heaters was 12kW. After 18 minutes of heating, the actual average temperature of the environmental chamber reached 157℃, or 85% of the set temperature, entering the mid-stage. At this point, the power ratio of each heater was adjusted in real time according to the temperature distribution of each temperature control zone. For temperature control zones with a temperature deviation greater than 5℃, the heater power was increased to 110% of the rated power; for temperature control zones with a temperature deviation between 2℃ and 5℃, the heater power remained at the rated power; and for temperature control zones with a temperature deviation less than 2℃, the heater power was reduced to 70% of the rated power. After 5 minutes of fine-tuning, the actual average temperature of the environmental chamber reached 176℃, or 95% of the set temperature, entering the late-stage. Technicians gradually reduced the number of low-power auxiliary heaters in operation and lowered the power of the high-power main heater. When the actual average temperature of the environmental chamber deviated from the set temperature by less than 5°C, one low-power auxiliary heater was shut down every 30 seconds until only two low-power auxiliary heaters remained in operation. At the same time, the power of the high-power main heater was reduced to 50% of its rated power for heat preservation control.

[0113] The Beckhoff CX2033 controller achieves precise power control of the high-power main heater and low-power auxiliary heater through the conduction angle adjustment of the thyristor power regulator, based on the optimal power allocation scheme. The conduction angle adjustment employs an adaptive step-size algorithm. The normalized temperature deviation is calculated as the absolute value of the current temperature deviation divided by the set temperature of 185°C, and the normalized temperature deviation change rate is the absolute value of the current temperature deviation change rate divided by 10°C per second. The conduction angle adjustment step value is a base step angle of 5 degrees multiplied by the square root of the sum of the normalized temperature deviation and the normalized temperature deviation change rate. When the absolute value of the temperature deviation is greater than 10°C or the absolute value of the temperature deviation change rate is greater than 2°C per second, a large step is used for rapid power adjustment. When the absolute value of the temperature deviation is less than 3°C and the absolute value of the temperature deviation change rate is less than 0.5°C per second, a small step is used for smooth transition to avoid temperature oscillations caused by power abrupt changes. Figure 5 As shown, the temperature uniformity of the environmental chamber was controlled at a low level throughout the entire heating test, and the temperature field distribution remained stable. Key temperature data during the test are shown in Table 1.

[0114] Table 1 Key Temperature Data for Heating Test

[0115]

[0116] This invention employs a graph theory-based topology optimization sensor placement algorithm to ensure sufficient observation capability of the temperature field map, avoiding redundancy or blind spots caused by traditional uniform placement methods, enabling sensors to accurately capture key features of the temperature field. A dual mechanism of sliding window statistical testing and Grubbs' test effectively eliminates abnormal temperature values, improving the reliability of calculating the actual average temperature of the environmental chamber. The fuzzy sliding mode variable structure control algorithm combines the robustness of sliding mode control with the smoothness of fuzzy control, providing rapid response under large deviations and smooth transition under small deviations, overcoming the performance imbalance of single control methods at different stages. The dynamic equilibrium optimization model of the thermal field uses a multi-head self-attention mechanism to model the coupling relationship between temperature control zones, dynamically adjusting the regularization intensity of each layer based on an adaptive weight decay parameter regularization framework, giving the model stronger generalization ability and adaptability when facing irregular test specimen shapes and different operating conditions. The variable power staged heating strategy dynamically adjusts the heater power distribution according to the heating process: initial full-power rapid heating to shorten the test time, fine adjustment in the middle to improve temperature distribution, and smooth transition in the later stage to ensure temperature stability, avoiding the slow heating rate or temperature overshoot problems caused by traditional fixed-power heating. The adaptive step size adjustment of the conduction angle of the thyristor power regulator dynamically determines the step value based on the temperature deviation value and the rate of change of temperature deviation. A large step is used when a fast response is required, and a small step is used when approaching the target temperature. This effectively prevents temperature oscillations caused by power step changes and improves the accuracy and stability of temperature control.

[0117] It should be noted that the variables involved in this invention are explained in detail in Tables 2 and 3.

[0118] Table 2. Variable Explanation Table (Part 1)

[0119]

[0120] Table 3. Variable Explanation Table (Part Two)

[0121]

[0122] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for environmental chamber heating tests, characterized in that, A temperature field topology sensor array is set up inside the environmental chamber. A graph theory topology optimization sensor placement algorithm is used to determine the installation positions of K-type thermocouple sensors. Real-time temperature data of all K-type thermocouple sensors in the temperature field topology sensor array are collected. After sliding window statistical verification to remove abnormal temperature values, the actual average temperature of the environmental chamber is calculated. A high-power main heater and a high-temperature resistant blower are installed at the top of the environmental chamber to form a closed-loop circulation system. Low-power auxiliary heaters are evenly distributed at the bottom of the environmental chamber. Based on the temperature deviation between the actual average temperature and the set temperature, the environmental chamber is divided into multiple temperature control zones. A fuzzy sliding mode variable structure control algorithm is used to calculate the zone heater control commands for each temperature control zone. The zone heater control commands of all temperature control zones are input into the thermal field dynamic balance optimization model for global collaborative optimization and output the optimal power allocation scheme for each heater. Based on the optimal power allocation scheme, the conduction angle of the thyristor power regulator is adjusted to achieve precise power control of the high-power main heater and the low-power auxiliary heater, and a variable power staged heating strategy is implemented.

2. The method according to claim 1, characterized in that, The graph theory-based topology optimization sensor placement algorithm specifically involves dividing the environmental chamber into three-dimensional node grids by equally dividing the length, width, and height directions. It calculates the spatial distance between any two temperature nodes and the correlation coefficient of historical temperature data, using the historical temperature data correlation coefficient as the weight value of the edges between temperature nodes to construct a temperature field map. K-means clustering analysis is then performed on the temperature field map to divide it into multiple sub-regions with strong temperature correlation. Within each sub-region, a greedy algorithm is used to solve for the minimum dominance set, selecting the minimum representative nodes that can cover all temperature nodes in the sub-region as the installation locations for the K-type thermocouple sensors.

3. The method according to claim 2, characterized in that, The correlation coefficient of the historical temperature data was calculated using Pearson correlation analysis, and the expected number of sensors was preset based on the volume of the environmental chamber at a density of 0.5 to 1 sensor per cubic meter.

4. The method according to claim 3, characterized in that, The sliding window statistical test process specifically involves establishing a sliding time window with a length of 30 sampling points, calculating the window mean and standard deviation for the real-time temperature data of each K-type thermocouple sensor within the sliding time window, performing a 3σ criterion test on the real-time temperature data of each K-type thermocouple sensor at the current sampling time to determine suspected abnormal temperature values, and further verifying the real-time temperature data of the K-type thermocouple sensors that are determined to be suspected abnormal temperature values ​​using the Grubbs test method.

5. The method according to claim 4, characterized in that, The Grubbs statistic is calculated by dividing the absolute value of the difference between the suspected abnormal temperature value and the sample mean within the sliding time window by the sample standard deviation within the sliding time window. The Grubbs critical value is obtained by referring to the Grubbs critical value table based on the sliding time window length of 30 and the significance level of 0.

05. The actual average temperature of the environmental chamber is obtained by removing the maximum and minimum temperature values ​​from all valid sensor temperature data after removing abnormal temperature values ​​and then performing an arithmetic mean.

6. The method according to claim 5, characterized in that, The fuzzy sliding mode variable structure control algorithm specifically defines the sliding mode switching function as a weighted sum of the temperature deviation value and the rate of change of the temperature deviation. When the absolute value of the sliding mode switching function is greater than the switching threshold, the sliding mode control law is used to calculate the control command for the partition heater. When the absolute value of the sliding mode switching function is less than or equal to the switching threshold, the algorithm switches to fuzzy control to calculate the control command for the partition heater.

7. The method according to claim 6, characterized in that, The weighting coefficient for the temperature deviation value in the sliding mode switching function is 7, the weighting coefficient for the rate of change of temperature deviation is 5, the switching threshold is 3℃, and the sliding mode gain coefficient is determined to be 15 based on the system uncertainty range and the upper limit of the disturbance.

8. The method according to claim 7, characterized in that, The fuzzy control establishes a fuzzy rule base for temperature deviation value and temperature deviation change rate. The fuzzy set of the fuzzy input variable of temperature deviation value is divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy set of the fuzzy input variable of temperature deviation change rate is divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The fuzzy set of the fuzzy output variable of power regulation is divided into 7 fuzzy levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large.

9. The method according to claim 8, characterized in that, The input layer of the thermal field dynamic balance optimization model receives four types of feature data: temperature deviation value of each temperature control zone, temperature deviation change rate, current power ratio of each heater, and actual average temperature of the environmental chamber. The first hidden layer contains 128 neurons using the ReLU activation function, the second hidden layer contains 64 neurons combined with a multi-head self-attention mechanism, and the third hidden layer contains 32 neurons. The number of neurons in the output layer is equal to the total number of heaters and uses the Softmax activation function.

10. The method according to claim 9, characterized in that, The dynamic equilibrium optimization model for the thermal field adopts a parameter regularization framework based on adaptive weight decay, calculating the gradient of parameters at each layer before each parameter update. Norm, will After normalizing the norm value to the interval of 0 to 1, multiply it by the basic attenuation coefficient to obtain the actual attenuation coefficient of the layer. In the parameter update formula, the parameter gradient is subtracted from the product of the parameter value and the actual attenuation coefficient.

Images