Multifunctional dual-channel testing device

Abstract

The invention relates to the technical field of heat dissipation testing, and discloses a multifunctional dual-channel testing device, which comprises an equipment main body rack used for providing an installation carrier; the dual-channel test module is mounted in the equipment main body rack and comprises two test channels which are independently arranged, and each test channel is provided with an independent heat source and a matched test pipeline; the multi-dimensional sensing module comprises a temperature sensing assembly of temperature sensing points capable of being flexibly deployed, a flow control assembly provided with flowmeters with different measuring ranges and a pressure sensing assembly comprising a water inlet pressure sensor and a pressure difference sensor; and each test channel is provided with a flow control assembly provided with flowmeters with different measuring ranges. Through the cooperation of the design of double independent channels and the expansion interface module, testing of liquid cooling and air cooling heat dissipation products can be carried out at the same time, performance evaluation of heat dissipation assemblies in multiple fields can be covered without additional equipment, the expansion interface can adapt to different jigs and test scenes, the product switching process is simplified, the test efficiency and universality are improved, and the test cost is reduced. And diversified test requirements can be met.

Description

Technical Field

[0001] This invention relates to the field of heat dissipation testing technology, specifically to a multifunctional dual-channel testing device. Background Technology

[0002] A heat dissipation testing device is a device used to evaluate the performance of various heat dissipation products. It can simulate the operating conditions of heat dissipation products in actual working environments, and assist in verifying the heat dissipation effect of heat dissipation products by monitoring key parameters related to heat dissipation. It is suitable for performance testing of different types of heat dissipation products and provides support for the optimization and verification of heat dissipation solutions.

[0003] In existing technologies, most testing devices are single-channel designs, making it difficult to conduct liquid cooling and air cooling tests simultaneously. If performance evaluation of both types of heat dissipation products needs to be covered, multiple dedicated devices are often required, which not only increases equipment investment costs but also occupies more space. Therefore, a multi-functional dual-channel testing device is proposed. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a multifunctional dual-channel testing device, which solves the problem that existing testing devices are mostly single-channel designs, making it difficult to conduct liquid cooling and air cooling tests simultaneously. If performance evaluation of two types of heat dissipation products is required, it increases equipment investment costs and occupies more space.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A multifunctional dual-channel testing device, comprising:

[0007] The main frame of the equipment serves as the mounting platform;

[0008] The dual-channel test module is installed inside the main frame of the equipment and includes two independently configured test channels. Each test channel is equipped with an independent heat source and matching test pipelines.

[0009] The multi-dimensional sensing module includes a temperature sensing component with flexibly deployable sensing points, a flow control component with flow meters of different ranges configured in each of the test channels, and a pressure sensing component including an inlet pressure sensor and a differential pressure sensor.

[0010] The temperature control module includes a constant temperature water tank with heating and cooling functions, which is connected to the test channel of the dual-channel test module through a circulation pipeline;

[0011] An expansion interface module is located on the side of the main frame of the equipment, including an air nozzle interface, a fan power supply interface, and a pressure sensor interface;

[0012] The intelligent control module connects the multi-dimensional sensing module and the temperature control module, and is used for predictive control through digital twin drive combined with PID control algorithm;

[0013] The collaborative detection module, associated with the multidimensional sensing module, is used for collaborative anomaly detection through federated learning.

[0014] Furthermore, the digital twin drive combined with the PID control algorithm to achieve predictive control includes the following steps:

[0015] A multiphysics digital twin model corresponding to the physical entity of the dual-channel test module is constructed. The model includes a fluid dynamics and heat transfer mechanism model and a PID control simulation unit.

[0016] Receive real-time sensing data from the multi-dimensional sensing module and dynamically calibrate the digital twin model and PID controller parameters based on data deviation;

[0017] When receiving a control task, the basic control quantity is first generated by performing calculations based on the PID regulation law.

[0018] The basic control variables are used in the digital twin model to accelerate simulation and predict the long-term, cross-channel coupling effects of control operations on the dual test channels.

[0019] Based on the prediction results and a multi-objective reward function that includes energy consumption, efficiency, and stability, the PID parameter combination is optimized through reinforcement learning algorithm to find the best option in the digital twin model and generate an optimized control command sequence.

[0020] The optimized control command sequence is sent to the flow control component and the constant temperature water tank, and the adjustment is performed through closed-loop PID control.

[0021] Furthermore, the multiphysics digital twin model is constructed by coupling the fluid continuity equation, the Navier-Stokes equation, and the Fourier heat conduction equation.

[0022] The dynamic calibration specifically involves inputting the residual between the real-time sensing data and the output value of the digital twin model into an extended Kalman filter to dynamically correct the boundary conditions and material property parameters of the model. At the same time, the proportional coefficient, integral time, and derivative time of the PID controller are adjusted based on the residual gradient.

[0023] Furthermore, the reinforcement learning algorithm is a proximal policy optimization algorithm, and the policy network output action space used to optimize the PID parameter combination is defined as a set of discrete values ​​of the adjustment range of the PID proportional coefficient, integral time, and derivative time.

[0024] Furthermore, in the multi-objective reward function, the weight of the energy consumption index is: The efficiency index has a weight of 1. The stability index has a weight of 1. ,in ,and , , The value is dynamically configured based on the current test task type.

[0025] Furthermore, the collaborative anomaly detection via federated learning includes the following steps:

[0026] Each test channel independently trains a local anomaly detection model using locally acquired multimodal data, which includes sensor time-series data, hardware status codes, environmental parameters, and historical PID control output data.

[0027] Without sharing the original data, the gradient and weight update parameters of the local anomaly detection model are encrypted and uploaded to the federated learning center server.

[0028] Receives a global anomaly detection model generated by aggregating update parameters from multiple clients from the Federated Learning Center server;

[0029] The global anomaly detection model is used to identify anomalies in the test process and output anomaly confidence scores. When an anomaly is detected, the emergency reset mechanism of PID parameters in the intelligent control module is triggered.

[0030] Furthermore, the local anomaly detection model adopts a hybrid architecture of temporal convolutional network and graph attention network, wherein the temporal convolutional network is used to extract the dynamic change features of the temporal data of PID control parameters;

[0031] The update parameters are encrypted using a homomorphic encryption algorithm;

[0032] The aggregation process of the global anomaly detection model adopts a weighted average algorithm based on model contribution, and the weights are dynamically adjusted according to the data coverage and annotation quality of each client.

[0033] Furthermore, in the weighted average algorithm based on model contribution, the client weights are dynamically adjusted according to the data coverage and annotation quality of each client. Specifically, the weight of each client is proportional to the product of its data coverage and annotation quality, and is normalized and allocated based on the sum of the products of all clients. Here, data coverage refers to the proportion of the number of data samples owned by the client to the total number of samples, and annotation quality refers to the score obtained by the central server based on the client's historical annotation accuracy.

[0034] Furthermore, the PID parameter emergency reset mechanism specifically restores the proportional coefficient, integral time, and derivative time of the current PID controller to the preset safety baseline parameter values ​​for this test phase stored in the intelligent control module.

[0035] Furthermore, the temporal convolutional network includes an LSTM model containing 128 hidden units, using the tanh activation function and the sigmoid gating function, with a time step of 60, to capture the long temporal dependency features of PID control parameters.

[0036] This invention provides a multifunctional dual-channel testing device. It has the following advantages:

[0037] 1. This invention, through the combination of dual independent channel design and expansion interface module, can simultaneously conduct tests on both liquid-cooled and air-cooled heat dissipation products. It can cover the performance evaluation of heat dissipation components in multiple fields without additional equipment. The expansion interface can be adapted to different fixtures and test scenarios, simplifying the product switching process, improving test efficiency and versatility, and helping to meet diverse test needs.

[0038] 2. The intelligent control module of this invention adopts digital twin drive combined with PID control algorithm. By constructing a multi-physics digital twin model and dynamically calibrating it, it can accurately predict the impact of control operation on the test channel. In addition, by combining reinforcement learning to optimize PID parameters, it can stabilize parameters such as temperature and flow rate within the target range, effectively improving test accuracy and efficiency and ensuring the stability of the test process.

[0039] 3. The collaborative detection module of this invention performs collaborative anomaly detection based on federated learning. Each test channel independently trains its model and encrypts and uploads parameters without sharing the original data. The aggregated global model can accurately identify local and cross-channel related anomalies. When an anomaly is detected, an emergency reset mechanism for PID parameters can be triggered to quickly restore test stability, which protects data privacy and improves the reliability of the test process. Attached Figure Description

[0040] Figure 1 This is a perspective view of the present invention;

[0041] Figure 2 This is a partial structural diagram of the main frame of the device of the present invention;

[0042] Figure 3 for Figure 2 Enlarged view of point A in the middle;

[0043] Figure 4 This is a partial structural diagram of the supporting test pipeline of the present invention;

[0044] Figure 5 This is a logic diagram of the test process of the present invention.

[0045] The components include: 1. Main frame of the equipment; 2. Circulation pipeline; 3. Constant temperature water tank; 4. Air nozzle interface; 5. Fan power supply interface; 6. Pressure sensor interface; 7. Temperature sensing components; and 8. Matching test pipelines. Detailed Implementation

[0046] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see the appendix Figure 1 -Appendix Figure 5 This invention provides a multifunctional dual-channel testing device, comprising:

[0048] The main frame 1 of the equipment serves as the mounting platform;

[0049] The dual-channel test module is installed in the main frame 1 of the equipment and includes two independently configured test channels. Each test channel is equipped with an independent heat source and a matching test pipeline 8.

[0050] The multi-dimensional sensing module includes a temperature sensing component 7 with flexibly deployable sensing points, a flow control component with flow meters of different ranges configured in each test channel, and a pressure sensing component containing an inlet pressure sensor and a differential pressure sensor.

[0051] The temperature control module includes a constant temperature water tank 3 with heating and cooling functions, which is connected to the test channel of the dual-channel test module through a circulation pipeline 2.

[0052] An expansion interface module is located on the side of the main frame 1 of the equipment, including an air nozzle interface 4, a fan power supply interface 5, and a pressure sensor interface 6.

[0053] Specifically, the testing process logic for this equipment is as follows:

[0054] Before testing, the liquid-cooled or air-cooled products to be tested are installed on the two independent test channels of the main frame 1 of the equipment. The air source is connected through the air nozzle interface 4 of the expansion interface module to ensure the product and the test fixture are firmly pressed together. When air-cooling test is required, the cooling fan is connected through the fan power supply interface 5 and the pressure sensor interface 6 is connected to the pressure monitoring element to ensure the stability of the pressing process.

[0055] Subsequently, the target temperature of the constant temperature water tank 3, the flow parameters of each channel flow control component and the heating power of the independent heat source are set according to the test requirements. After the equipment is started, the constant temperature water tank 3 delivers a stable circulating medium to the test channel through the circulation pipeline 2. At the same time, the independent heat source starts to simulate the heating of the device, forming a test environment consistent with the actual working conditions.

[0056] During the test, the multi-dimensional sensing module monitors the entire process. The temperature sensing component 7 collects the temperature of key parts of the product and the inlet and outlet temperatures of the medium in real time. The flow control component regulates and feeds back the actual flow rate. The inlet water pressure sensor and differential pressure sensor monitor the pipeline pressure status. The data is synchronously transmitted to the control system for processing and analysis.

[0057] After the test is completed, the complete data curve is automatically recorded, and a test report is generated as needed;

[0058] The main frame 1 of the equipment has two independent channels in the dual-channel test module. Each channel is equipped with an independent heat source to simulate the heating conditions of different power devices. The matching test pipeline 8 is used to provide a medium flow path for liquid cooling test. During the test, the peripheral settings are first completed through the expansion interface module. The air nozzle interface 4 is connected to the external air source to connect to the special fixture to achieve stable pressing of the test product. The fan power supply interface 5 provides power to the cooling fan in the air-cooled test scenario.

[0059] In the temperature control process, the constant temperature water tank 3, which has heating and cooling functions, delivers temperature-controlled fluid medium to the two test channels through the circulation pipeline 2, forming a closed-loop temperature control system. This provides the basic conditions for heat dissipation performance testing under different ambient temperatures. During the test, the temperature sensing component 7, with its flexibly deployable sensing points, captures the temperature field distribution of key parts of the test product and the fluid inlet and outlet. The flow control component, composed of flow meters with different ranges equipped in each channel, can adjust and monitor the fluid flow according to the test requirements, adapting to the test requirements of products with different heat dissipation power. The inlet pressure sensor and differential pressure sensor are used to monitor pipeline pressure parameters to ensure stable test conditions.

[0060] The parallel design with dual independent channels supports simultaneous testing of two different types of products or comparative testing of the same product. Distributed temperature measurement and multi-parameter collaborative acquisition ensure the integrity of test data, and the modular interface design simplifies the switching process for different types of products, which helps to improve testing efficiency and versatility.

[0061] The intelligent control module connects the multi-dimensional sensing module and the temperature control module, and is used for predictive control through digital twin drive combined with PID control algorithm;

[0062] Furthermore, the digital twin drive combined with PID control algorithm to achieve predictive control includes the following steps:

[0063] A multiphysics digital twin model corresponding to the physical entity of the dual-channel test module is constructed. The model includes fluid dynamics and heat transfer mechanism models and PID control simulation units.

[0064] Receive real-time sensing data from the multi-dimensional sensing module and dynamically calibrate the digital twin model and PID controller parameters based on data deviation;

[0065] When receiving a control task, the basic control quantity is first generated by performing calculations based on the PID regulation law.

[0066] Accelerate simulation by using basic control variables in the digital twin model to predict the long-term, cross-channel coupling effects of control operations on the dual test channels;

[0067] Based on the prediction results and a multi-objective reward function that includes energy consumption, efficiency, and stability, the PID parameter combination is optimized through reinforcement learning algorithm, and the optimal control command sequence is generated in the digital twin model.

[0068] The optimized control command sequence is sent to the flow control component and the constant temperature water tank 3, and the adjustment is performed through closed-loop PID control.

[0069] Furthermore, the multiphysics digital twin model is constructed by coupling the fluid continuity equation, the Navier-Stokes equation, and the Fourier heat conduction equation.

[0070] Dynamic calibration specifically involves inputting the residual between real-time sensor data and the output value of the digital twin model into an extended Kalman filter to dynamically correct the boundary conditions and material property parameters of the model. At the same time, the proportional coefficient, integral time, and derivative time of the PID controller are adjusted based on the residual gradient.

[0071] Furthermore, the reinforcement learning algorithm is a proximal policy optimization algorithm, and the policy network output action space used to optimize the PID parameter combination is defined as the set of discrete values ​​of the adjustment amplitude of the PID proportional coefficient, integral time and derivative time.

[0072] Furthermore, in the multi-objective reward function, the weight of the energy consumption index is... The efficiency index has a weight of 1. The stability index has a weight of 1. ,in ,and , , The value is dynamically configured based on the current test task type;

[0073] Specifically, the intelligent control module connects the multi-dimensional sensing module and the temperature control module. It achieves predictive control through digital twin drive combined with PID control algorithm. When constructing a multi-physics digital twin model, the fluid continuity equation is coupled to describe the mass conservation of the fluid in the test channel, the Navier-Stokes equation reflects the change of fluid momentum, and the Fourier heat conduction equation simulates the heat transfer process. At the same time, the PID control simulation unit is integrated to reproduce the control logic. During the model operation, the inlet and outlet temperature, real-time flow rate and pressure data of the test channel collected by the multi-dimensional sensing module are continuously received. The residuals of these data and the model output values ​​are input into the extended Kalman filter to dynamically correct the boundary conditions of the model, such as pipe roughness and material property parameters such as thermal conductivity. At the same time, the proportional coefficient, integral time and derivative time of the PID controller are adjusted based on the residual gradient to make the model output consistent with the actual test state.

[0074] When receiving a control task, a basic control quantity is first generated based on the PID control law, where the PID calculation formula is:

[0075] ;

[0076] In the formula This represents the deviation between the target temperature and the actual temperature. This is the proportionality coefficient. For integration time, The differential time is used to obtain the initial flow regulation and temperature control quantities. The basic control quantities are then input into the digital twin model for accelerated simulation to simulate the flow field distribution and temperature field changes of the dual test channels under different control operations. This helps predict the cross-channel coupling effects over long time, such as the interference of flow rate changes in one channel on the temperature stability of another channel.

[0077] Based on simulation prediction results and a multi-objective reward function, the PID parameter combination is optimized using a reinforcement learning algorithm. The reward function formula is:

[0078] ;

[0079] in , It is an energy consumption indicator and is positively correlated with the heating and cooling power of the constant temperature water tank. It is an efficiency indicator and is positively correlated with flow control accuracy. It is a stability indicator and is negatively correlated with the amplitude of temperature fluctuations. , , The weight coefficients are dynamically configured according to the test task type, such as water-cooled plate testing or air-cooled heat sink testing. Reinforcement learning adopts a near-end policy optimization algorithm. The policy network output action space is a set of discrete values ​​of the adjustment amplitude of PID parameters. An optimized control command sequence is generated by iterative optimization in the digital twin model. This sequence is sent to the flow control component and the constant temperature water tank 3. Closed-loop PID control keeps the temperature, flow rate and other parameters within the target range during the test, improving the test accuracy and efficiency.

[0080] The collaborative detection module, associated with the multi-dimensional sensing module, is used for collaborative anomaly detection through federated learning.

[0081] Furthermore, collaborative anomaly detection through federated learning includes the following steps:

[0082] Each test channel independently trains a local anomaly detection model using locally acquired multimodal data, which includes sensor time-series data, hardware status codes, environmental parameters, and historical PID control output data.

[0083] Without sharing the original data, the gradient and weight update parameters of the local anomaly detection model are encrypted and uploaded to the federated learning center server.

[0084] Receives a global anomaly detection model generated by aggregating update parameters from multiple clients from the Federated Learning Center server;

[0085] A global anomaly detection model is used to identify anomalies in the test process and output anomaly confidence scores. When an anomaly is detected, the emergency reset mechanism of PID parameters in the intelligent control module is triggered.

[0086] Furthermore, the local anomaly detection model adopts a hybrid architecture of temporal convolutional network and graph attention network, where the temporal convolutional network is used to extract the dynamic change features of the time series data of PID control parameters;

[0087] The updated parameters are encrypted using a homomorphic encryption algorithm;

[0088] The aggregation process of the global anomaly detection model adopts a weighted average algorithm based on model contribution, with the weights dynamically adjusted according to the data coverage and annotation quality of each client.

[0089] Furthermore, in the weighted average algorithm based on model contribution, the client weights are dynamically adjusted according to the data coverage and annotation quality of each client. Specifically, the weight of each client is proportional to the product of its data coverage and annotation quality, and is normalized based on the sum of the products of all clients. Here, data coverage refers to the proportion of the number of data samples owned by the client to the total number of samples, and annotation quality refers to the score obtained by the central server based on the client's historical annotation accuracy.

[0090] Furthermore, the PID parameter emergency reset mechanism specifically restores the proportional coefficient, integral time, and derivative time of the current PID controller to the preset safety baseline parameter values ​​stored in the intelligent control module for this test phase.

[0091] Furthermore, the temporal convolutional network includes an LSTM model containing 128 hidden units, using the tanh activation function and the sigmoid gating function, with a time step of 60, to capture the long temporal dependency features of PID control parameters;

[0092] Specifically, the collaborative detection module is linked to the multi-dimensional sensing module. The collaborative anomaly detection process through federated learning begins with local model training. Each test channel collects multimodal data to train the local anomaly detection model. This multimodal data includes sensor time-series data such as temperature change sequences, flow fluctuation curves, and pressure difference time-series data at the inlet and outlet of the test channel; hardware status codes such as equipment operation stage identifiers and component status codes; environmental parameters such as room temperature and humidity in the test environment; and historical PID control output data such as real-time adjustment values ​​of proportional coefficient, integral time, and derivative time. The local anomaly detection model adopts a hybrid architecture of temporal convolutional networks and graph attention networks, with the core of the temporal convolutional network being an LSTM model. The model contains 128 hidden units to enhance feature extraction capabilities. It uses the tanh activation function to handle nonlinear changes in cell state and the sigmoid gating function, including input gate, forget gate, and output gate, to precisely control the inflow, retention, and output of information. It sets 60 time steps to cover the dynamic changes of PID control parameters within a 1-minute cycle, thereby effectively capturing long-term time-series dependencies between parameters. The graph attention network uses each modal data as a node and learns the correlation weights between modalities, such as the correlation strength between the rate of temperature change and the adjustment amplitude of PID integral time, through the attention mechanism. After the features of the two are fused, they are input into the fully connected layer to output anomaly probability values ​​to reflect the degree of anomaly in the current test state.

[0093] After local model training is complete, the federated learning collaborative phase begins. Each test channel, without sharing the original data, uses a homomorphic encryption algorithm to encrypt the gradients of its local model, such as the weight gradients of convolutional layers and the bias gradients of fully connected layers, as well as the weight update parameters. The encrypted parameters are then uploaded to the federated learning center server. The server aggregates the encrypted parameters from each client and calculates the global parameters using a weighted average algorithm based on model contribution. The client weight formula is as follows: ,in For client weight, Data coverage refers to the proportion of valid samples held by the client out of the total number of samples. The quality of annotation is determined by the score obtained by the central server based on the accuracy of the client's historical annotation data. The algorithm generates a global anomaly detection model and distributes it to each client to achieve collaborative optimization of the model. This preserves local data privacy while aggregating multi-source knowledge to improve detection generalization ability.

[0094] After deployment, the global anomaly detection model runs in real-time during the testing process. It continuously receives real-time data from the multi-dimensional sensing module and outputs anomaly confidence scores to quantify the degree of anomaly. When the score reaches a set threshold, the system determines it as an anomaly and triggers the emergency reset mechanism of the PID parameters in the intelligent control module. This restores the proportional coefficient integral time and derivative time of the current PID controller to the preset safety baseline parameter values ​​stored in the intelligent control module for this testing phase. These baseline parameters are statistically determined based on a large number of normal test PID parameters, enabling rapid stabilization of the test state. The LSTM model expands its capabilities through parallel computation of 128 hidden units. In terms of feature dimension, the tanh function normalizes the input data, compressing the values ​​to the [-1,1] interval to avoid gradient explosion. The sigmoid gating function dynamically adjusts the importance of information at each time step through the output value in the 0-1 interval. The sliding window with a time step length of 60 ensures that the model can capture the trend changes of PID parameters in typical test cycles, thereby accurately identifying abnormal patterns in long-term time-series dependencies. Combined with the collaborative mechanism of federated learning, it can accurately identify local anomalies such as single-channel flow anomalies, and also perceive cross-channel correlation anomalies such as two-channel pressure coupling anomalies, improving the stability and reliability of the testing process.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multifunctional dual-channel testing device, characterized in that, include: The main frame of the equipment (1) is used to provide an installation carrier; The dual-channel test module is installed in the main frame (1) of the equipment and includes two independently set test channels. Each test channel is equipped with an independent heat source and a matching test pipeline (8). The multidimensional sensing module includes a temperature sensing component (7) with flexibly deployable sensing points, a flow control component with flow meters of different ranges configured in each of the test channels, and a pressure sensing component including an inlet pressure sensor and a differential pressure sensor. The temperature control module includes a constant temperature water tank (3) with heating and cooling functions, which is connected to the test channel of the dual-channel test module through a circulation pipeline (2); An expansion interface module is located on the side of the main frame (1) of the equipment, including an air nozzle interface (4), a fan power supply interface (5) and a pressure sensor interface (6). The intelligent control module connects the multi-dimensional sensing module and the temperature control module, and is used for predictive control through digital twin drive combined with PID control algorithm; The collaborative detection module, associated with the multidimensional sensing module, is used for collaborative anomaly detection through federated learning.

2. The multifunctional dual-channel testing device according to claim 1, characterized in that, The digital twin drive combined with the PID control algorithm to achieve predictive control includes the following steps: A multiphysics digital twin model corresponding to the physical entity of the dual-channel test module is constructed. The model includes a fluid dynamics and heat transfer mechanism model and a PID control simulation unit. Receive real-time sensing data from the multi-dimensional sensing module and dynamically calibrate the digital twin model and PID controller parameters based on data deviation; When receiving a control task, the basic control quantity is first generated by performing calculations based on the PID regulation law. The basic control variables are used in the digital twin model to accelerate simulation and predict the long-term, cross-channel coupling effects of control operations on the dual test channels. Based on the prediction results and a multi-objective reward function that includes energy consumption, efficiency, and stability, the PID parameter combination is optimized through reinforcement learning algorithm to find the best option in the digital twin model and generate an optimized control command sequence. The optimized control command sequence is sent to the flow control component and the constant temperature water tank (3) and the adjustment is performed through closed-loop PID control.

3. The multifunctional dual-channel testing device according to claim 2, characterized in that, The multiphysics digital twin model is constructed by coupling the fluid continuity equation, the Navier-Stokes equation, and the Fourier heat conduction equation. The dynamic calibration specifically involves inputting the residual between the real-time sensing data and the output value of the digital twin model into an extended Kalman filter to dynamically correct the boundary conditions and material property parameters of the model. At the same time, the proportional coefficient, integral time, and derivative time of the PID controller are adjusted based on the residual gradient.

4. The multifunctional dual-channel testing device according to claim 2, characterized in that, The reinforcement learning algorithm is a near-end policy optimization algorithm. The policy network output action space used to optimize the PID parameter combination is defined as a set of discrete values ​​of the adjustment range of the PID proportional coefficient, integral time, and derivative time.

5. A multifunctional dual-channel testing device according to claim 2, characterized in that, In the multi-objective reward function, the weight of the energy consumption index is: The efficiency index has a weight of 1. The stability index has a weight of 1. ,in ,and , , The value is dynamically configured based on the current test task type.

6. The multifunctional dual-channel testing device according to claim 1, characterized in that, The collaborative anomaly detection via federated learning includes the following steps: Each test channel independently trains a local anomaly detection model using locally acquired multimodal data, which includes sensor time-series data, hardware status codes, environmental parameters, and historical PID control output data. Without sharing the original data, the gradient and weight update parameters of the local anomaly detection model are encrypted and uploaded to the federated learning center server. Receives a global anomaly detection model generated by aggregating update parameters from multiple clients from the Federated Learning Center server; The global anomaly detection model is used to identify anomalies in the test process and output anomaly confidence scores. When an anomaly is detected, the emergency reset mechanism of PID parameters in the intelligent control module is triggered.

7. A multifunctional dual-channel testing device according to claim 6, characterized in that, The local anomaly detection model adopts a hybrid architecture of temporal convolutional network and graph attention network, wherein the temporal convolutional network is used to extract the dynamic change features of PID control parameter time series data; The update parameters are encrypted using a homomorphic encryption algorithm; The aggregation process of the global anomaly detection model adopts a weighted average algorithm based on model contribution, and the weights are dynamically adjusted according to the data coverage and annotation quality of each client.

8. A multifunctional dual-channel testing device according to claim 7, characterized in that, In the weighted average algorithm based on model contribution, the client weights are dynamically adjusted according to the data coverage and annotation quality of each client. Specifically, the weight of each client is proportional to the product of its data coverage and annotation quality, and is normalized based on the sum of the products of all clients. Here, data coverage refers to the proportion of the number of data samples owned by the client to the total number of samples, and annotation quality refers to the score obtained by the central server based on the client's historical annotation accuracy.

9. A multifunctional dual-channel testing device according to claim 6, characterized in that, The PID parameter emergency reset mechanism specifically restores the proportional coefficient, integral time, and derivative time of the current PID controller to the preset safety baseline parameter values ​​for this test phase stored in the intelligent control module.

10. A multifunctional dual-channel testing device according to claim 7, characterized in that, The temporal convolutional network includes an LSTM model with 128 hidden units, using the tanh activation function and the sigmoid gating function, with a time step of 60, and is used to capture the long temporal dependency features of PID control parameters.

Images