Sensor intelligent layout method and system in phase change direct cooling system

Abstract

The application discloses a kind of sensor intelligent distribution methods and systems in phase change direct cooling system, belong to electric digital data processing technical field, the method includes:1. structure key area definition;2. based on the classification of measured thermal characteristic area of infrared thermal image scanning;3. based on the fusion and optimization of structure key area and measured thermal characteristic area;By using genetic algorithm to solve iteratively, and after each round population iteration, call penalty function to complete constraint condition check, evolve the optimal distribution coordinates scheme under the condition of given sensor quantity.The application can realize the limited sensor arrangement at the best point by combining cold plate design, measured thermal image and optimization algorithm, so as to maximize the capture of cold plate temperature field key features, solve the technical problems of high cost, complex test process, inaccurate test results and other technical problems caused by test condition restriction or too many test points in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of electronic digital data processing technology, specifically relating to a method and system for intelligent sensor placement in a phase change direct cooling system. Background Technology

[0002] Phase change direct cooling technology utilizes the principle that liquid refrigerant absorbs a large amount of latent heat of vaporization during the evaporation phase change process to achieve efficient heat exchange. It is gaining widespread application in the battery thermal management systems of energy storage power stations and new energy vehicles. Its core principle is to bring the liquid refrigerant within the cold plate into contact with the heat source. For example, using environmentally friendly refrigerants such as R410A to contact the bottom of the battery cell or module, the refrigerant evaporates and absorbs heat in situ within the cold plate's flow channels, thus eliminating the secondary heat exchange step of the coolant in traditional liquid cooling solutions. Therefore, phase change direct cooling systems have significant advantages such as low thermal resistance, high energy efficiency, and fast temperature response.

[0003] As a key component for heat transfer in phase change direct cooling systems, the cold plate's structural design and surface temperature uniformity directly determine the temperature difference control capability within the battery module. To evaluate the performance of the cold plate, multiple temperature sensors, such as thermocouples, are typically placed on its surface. By collecting temperature values ​​at each measuring point, the uniformity of temperature distribution on the cold plate surface can be quantitatively analyzed. Traditional sensor placement methods rely heavily on engineers' experience, often employing a uniform grid or placing sensors at specific locations based on symmetry. However, these methods have the following inherent drawbacks: 1. Uniform or symmetrical point distribution may fail to fully consider factors that play a decisive role in the temperature field distribution, such as the internal flow channel structure and liquid distribution structure of the cold plate. This may cause the temperature sensor to miss key hot spots or high temperature gradient areas, making the test results unable to truly reflect the worst-case operating conditions of the cold plate.

[0004] 2. To ensure the detection of potential anomalies, there is a tendency to deploy an excessive number of temperature sensors. This not only increases testing costs and wiring complexity but also burdens data processing. However, many of these temperature sensors may be located in areas with gentle temperature changes and low data volume, resulting in limited contribution of their data to performance evaluation and a waste of resources.

[0005] 3. For cold-rolled steel plates of the same design, slight differences in manufacturing processes, such as welding uniformity and microchannel blockage, may result in deviations in the actual temperature field distribution during operation compared to the theoretical design. Fixed empirical point layout schemes cannot dynamically adapt to these individual differences, leading to deviations in the testing and evaluation of certain individual cold-rolled steel plates.

[0006] Currently, existing technologies lack a systematic method that combines cold plate design knowledge, measured thermal characteristics, and mathematical optimization objectives. This makes it difficult to objectively and quantitatively determine the optimal sensor placement scheme under a given constraint on the number of temperature sensors, thus hindering the acquisition of the maximum amount of information with the fewest sensors. Therefore, it is necessary to provide a scientific, efficient, and adaptive intelligent sensor placement technology to maximize the capture of key characteristics of the cold plate temperature field with limited temperature sensor resources, thereby improving the accuracy and efficiency of testing. Summary of the Invention

[0007] To address the aforementioned technical problems in existing technologies, this invention proposes a method and system for intelligent sensor placement in a phase change direct cooling system. By combining cold plate design, measured thermal imaging, and optimization algorithms, this invention can place a limited number of sensors at optimal locations, thereby maximizing the capture of key features of the cold plate temperature field. This solves the technical problems of high cost, complex testing procedures, and inaccurate test results caused by limitations in testing conditions or too many testing points in existing technologies.

[0008] A method for intelligent sensor placement in a phase change direct cooling system includes the following steps: S1. Definition of critical structural regions; S2. Classification of measured thermal feature regions based on infrared thermal imaging scanning; Under typical steady-state conditions, an infrared thermal imager is used to scan the lower surface of the cold plate to obtain a thermal feature map reflecting the actual working state of the cold plate; through thermal feature extraction and classification, the measured thermal feature areas in the thermal feature map are identified. S3. Integration and optimization based on the key structural region and the measured thermal characteristic region; The key structural region and the measured thermal characteristic region are logically superimposed, and weights are assigned to the superimposed region to generate a two-dimensional weight distribution map that reflects the comprehensive information value of the expected and measured information at each location. Points are placed in the superimposed regions on the two-dimensional weight distribution map, and the number of points in each region is adjusted according to the weight. The objective function for all sensor placements is established and solved iteratively using a genetic algorithm. After each round of population iteration, a penalty function is called to verify the constraints. Through selection, crossover, mutation operations and penalty constraint verification during the iteration process, the optimal sensor placement coordinate scheme under the given number of sensors is evolved.

[0009] In S1, the method for defining the critical area of ​​the structure is as follows: analyze the cold plate design data and historical layout experience data, and define the critical area of ​​the cold plate based on the mass-heat coupling law of phase change heat transfer, the hydraulic characteristics of fluid flow and the heat conduction mechanism of the cold plate structure.

[0010] In S1, the defined key structural regions include the liquid distribution chamber and cold plate inlet region, the parallel flow channel core heat exchange region, the inter-flow channel fin heat conduction region, the manifold and flow channel outlet region, and the geometric symmetry axis region.

[0011] In S2, a typical steady-state operating condition refers to: placing a battery on the upper surface of the cold plate, with the battery completely covering all flow channel areas within the cold plate, allowing the battery to charge and discharge under constant power conditions, with a charging and discharging operation time of not less than 1 hour, and the process of charging the battery from 0% to 100% or discharging it from 100% to 0% after at least 180 seconds of startup.

[0012] In S3, the method for logically superimposing the structural critical area and the measured thermal characteristic area is as follows: the structural critical area is projected onto the physical coordinates of the cold plate surface, and the logical superposition of the structural critical area and the measured thermal characteristic area is realized in the same physical coordinate system.

[0013] In S3, the superimposed region includes an overlapping region where the structural critical region and the measured thermal characteristic region overlap, a first non-overlapping region that includes only the measured thermal characteristic region, a second non-overlapping region that includes only the structural critical region, and a third non-overlapping region that includes both the non-structural critical region and the non-measured thermal characteristic region.

[0014] In S3, the method for assigning weights to the superimposed regions is as follows: assign a first-level weight to the overlapping region, and assign a second-level weight, a third-level weight, and a fourth-level weight to the first non-overlapping region, the second non-overlapping region, and the third non-overlapping region, respectively, with the values ​​of the first-level weight, the second-level weight, the third-level weight, and the fourth-level weight decreasing sequentially.

[0015] In S3, the specific process of iteratively solving using a genetic algorithm is as follows: S3.1, Establish the calculation foundation: Set the number of points to n, establish a coordinate system with any point on the cold plate as the origin, determine the coordinates of the n points, and normalize the coordinates of all points to integers; S3.2, Encoding Processing: The coordinates of each point are converted into a binary encoded string using a genetic algorithm. S3.3, Generate the initial population: Given the number of sensors z, randomly generate z initial individuals to form the initial population. Each initial individual corresponds to a set of z placement schemes randomly selected from n preset placement points, and each initial individual corresponds to a unique binary code string, which serves as the initial basis for iteratively selecting the optimal placement point. S3.4, Perform iterative operations: Perform multiple rounds of iteration on the initial population. In each round of iteration, the three core operations of selection, crossover, and mutation are completed sequentially. In the selection operation, each round of iteration first uses the roulette wheel method to select excellent individuals for crossover. The crossover operation adopts single-point crossover or two-point crossover, which is achieved by randomly exchanging binary code string fragments of different initial individuals in the initial population, thereby completing the combination optimization of the placement scheme. In the mutation operation, the mutation probability P is set. Each mutation is within the preset mutation range C. The binary code string corresponding to the new individual generated after the initial individual is exchanged is randomly flipped to realize the random adjustment of the z placement points and gradually optimize the placement selection results. S3.5, Iterative Selection and Convergence Judgment: In each iteration, the probability I of each individual being selected is calculated using the roulette wheel method. Based on the calculated probability I, z individuals are randomly selected. At the same time, a preset error e and the number of iterations are set to 20~50. When the iteration error of the largest fitness value among all individuals is less than the preset error e in each iteration within the set number of iterations, convergence is achieved. In S3.5, the probability I of each individual being selected is calculated using the following formula: Where i = 1, 2, ..., z; In the formula, I The probability of an individual being selected. For the first i The set of z placement points corresponding to each individual; This is the fitness value.

[0016] S3.6, After each iteration, a penalty function is called for verification: The penalty function uses the Euclidean distance method for verification. A minimum distance amin is set, and the distance between each placement point and any three other placement points is calculated in turn. The minimum distance is taken, and the average of the minimum distances of z placement points is taken as a1. If a1 < amin, it is judged as too dense, and this individual is eliminated. If the new individual after the iteration does not coincide with the set placement point on the coordinates, the straight-line distance a2 between the placement point and the nearest set placement point is calculated. If a2 < amin, the placement point is replaced by the nearest set placement point. If a2 > amin, it is judged as unqualified, the individual is eliminated, and a new individual is randomly introduced for the next iteration. S3.7, Output Results: After the genetic algorithm converges iteratively, output the optimal point layout coordinate scheme for the z sensors.

[0017] A smart sensor deployment system for a phase change direct cooling system includes the following steps: The structural critical area definition module is used to define the structural critical area of ​​the cold plate based on the cold plate design data and historical layout experience data, and according to the mass-heat coupling law of phase change heat transfer, the hydraulic characteristics of fluid flow and the heat conduction mechanism of the cold plate structure. The measured thermal feature area classification module is used to scan the lower surface of the cold plate with an infrared thermal imager under typical steady-state conditions to obtain thermal feature maps that reflect the actual working state of the cold plate; and to identify the measured thermal feature areas in the thermal feature maps through thermal feature extraction and classification. The fusion and optimization module is used to logically overlay the key structural region with the measured thermal feature region, and assign weights to the overlaid region to generate a two-dimensional weight distribution map that reflects the comprehensive information value of the expected and measured information at each location. It is used to place points on the overlaid region on the two-dimensional weight distribution map and adjust the number of points in each region according to the weights. It is used to establish the objective function for all points, use a genetic algorithm to solve iteratively, and call a penalty function after each round of population iteration to complete the constraint verification. Through selection, crossover, mutation operations and penalty constraint verification in the iteration process, the optimal point coordinate scheme under the given number of sensors is evolved.

[0018] In the fusion and optimization module, the method for logically superimposing the structural key area and the measured thermal characteristic area is as follows: the structural key area is projected onto the physical coordinates of the cold plate surface, and the logical superposition of the structural key area and the measured thermal characteristic area is realized in the same physical coordinate system.

[0019] In the fusion and optimization module, the superimposed region includes an overlapping region where the structural critical region and the measured thermal characteristic region overlap, a first non-overlapping region that includes only the measured thermal characteristic region, a second non-overlapping region that includes only the structural critical region, and a third non-overlapping region that includes both the non-structural critical region and the non-measured thermal characteristic region.

[0020] In the fusion and optimization module, the method for assigning weights to the superimposed regions is as follows: a first-level weight is assigned to the overlapping region, and second-level, third-level, and fourth-level weights are assigned to the first, second, and third non-overlapping regions, respectively, with the values ​​of the first-level, second-level, third-level, and fourth-level weights decreasing sequentially.

[0021] The specific process of iterative solution using a genetic algorithm in the fusion and optimization module is as follows: S3.1, Establish the calculation foundation: Set the number of points to n, establish a coordinate system with any point on the cold plate as the origin, determine the coordinates of the n points, and normalize the coordinates of all points to integers; S3.2, Encoding Processing: The coordinates of each point are converted into a binary encoded string using a genetic algorithm. S3.3, Generate the initial population: Given the number of sensors z, randomly generate z initial individuals to form the initial population. Each initial individual corresponds to a set of z placement schemes randomly selected from n preset placement points, and each initial individual corresponds to a unique binary code string, which serves as the initial basis for iteratively selecting the optimal placement point. S3.4, Perform iterative operations: Perform multiple rounds of iteration on the initial population. In each round of iteration, the three core operations of selection, crossover, and mutation are completed sequentially. In the selection operation, each round of iteration first uses the roulette wheel method to select excellent individuals for crossover. The crossover operation adopts single-point crossover or two-point crossover, which is achieved by randomly exchanging binary code string fragments of different initial individuals in the initial population, thereby completing the combination optimization of the placement scheme. In the mutation operation, the mutation probability P is set. Each mutation is within the preset mutation range C. The binary code string corresponding to the new individual generated after the initial individual is exchanged is randomly flipped to realize the random adjustment of the z placement points and gradually optimize the placement selection results. S3.5, Iterative Selection and Convergence Judgment: In each iteration, the probability I of each individual being selected is calculated using the roulette wheel method. Based on the calculated probability I, z individuals are randomly selected. At the same time, a preset error e and the number of iterations are set to 20~50. When the iteration error of the largest fitness value among all individuals is less than the preset error e in each iteration within the set number of iterations, convergence is achieved. In S3.5, the probability I of each individual being selected is calculated using the following formula: Where i = 1, 2, ..., z; In the formula, I The probability of an individual being selected. For the first i The set of z placement points corresponding to each individual; This is the fitness value.

[0022] S3.6, After each iteration, a penalty function is called for verification: The penalty function uses the Euclidean distance method for verification. A minimum distance amin is set, and the distance between each placement point and any three other placement points is calculated in turn. The minimum distance is taken, and the average of the minimum distances of z placement points is taken as a1. If a1 < amin, it is judged as too dense, and this individual is eliminated. If the new individual after the iteration does not coincide with the set placement point on the coordinates, the straight-line distance a2 between the placement point and the nearest set placement point is calculated. If a2 < amin, the placement point is replaced by the nearest set placement point. If a2 > amin, it is judged as unqualified, the individual is eliminated, and a new individual is randomly introduced for the next iteration. S3.7, Output Results: After the genetic algorithm converges iteratively, output the optimal point layout coordinate scheme for the z sensors.

[0023] Compared with the prior art, the beneficial effects of the present invention are: 1. The placement of the points in this invention represents the optimal balance between the critical structural area and the measured thermal characteristic area within a mathematical optimization framework. The overall method integrates the critical structural area and infrared thermographic data, and utilizes optimization algorithms to enable a limited number of sensors to accurately cover key design points and temperature characteristic areas. This allows for the most comprehensive temperature field characterization with the fewest measurement points, significantly improving testing efficiency and diagnostic accuracy.

[0024] Furthermore, in step S3, genetic algorithm features are employed. Innovatively, integer regularization preprocessing of cold plate placement coordinates narrows the algorithm's search range and reduces computational load. Binary encoding is then used to adapt the genetic operator's operational logic to improve iteration efficiency. Multiple random placement schemes are generated to construct a diverse initial population, broadening the global optimization range and improving solution accuracy from the source. The iteration process employs a multi-level optimization strategy: roulette wheel selection, single-point or two-point crossover recombination, and low-probability controlled mutation within a preset interval. This strategy preserves high-quality placement schemes, accelerates algorithm convergence, and... By moving beyond local optima, the effectiveness of the sensor placement scheme is ensured. Simultaneously, by combining roulette wheel probability selection with a dual convergence criterion of iteration count and fitness error, the accuracy and stability of the optimization results are guaranteed while controlling computational power consumption. An innovative penalty function based on Euclidean distance is introduced to perform dual verification of the spacing and coordinate compliance of individual sensors in each iteration. Invalid individuals with excessively dense placement or out-of-constraint locations are eliminated, and new individuals are added, effectively avoiding problems such as sensor clustering and layout failure. Ultimately, an optimal sensor placement scheme that combines the rationality of heat exchange monitoring with engineering feasibility is obtained.

[0025] 2. By integrating design mechanisms and measured data, this invention transforms sensor placement from experience-driven to data- and experience-driven, ensuring that sensors are preferentially deployed in locations most likely to experience problems and are richest in information. This allows for more reliable assessment of the temperature uniformity limits of cold plates and capture of abnormal operating conditions with a limited number of sensors, significantly improving the effectiveness of cold plate testing.

[0026] 3. Under the premise of achieving the same or even better test results, this invention can usually reduce the number of sensors required, or obtain more effective information using the same number of sensors. This not only reduces the material cost, wiring complexity and channel requirements of the data acquisition system for a single test, but also improves the reusability and test efficiency of the test platform.

[0027] 4. By introducing an infrared thermal imaging scanning process, this invention enables the sampling scheme to be dynamically adjusted according to the actual manufacturing and working performance of each cold plate, thus possessing the ability to adapt to individual differences. This overcomes the shortcomings of fixed schemes that cannot dynamically adapt to individual product differences, making the testing and evaluation more fair and accurate.

[0028] 5. This invention provides a clear and operable standardized test site layout process, reducing human arbitrariness and ensuring consistency and comparability of test site layouts for different batches and personnel. It provides accurate and upfront data support for the research and development, process optimization, and selection of cold plates, which cannot be obtained by traditional whole-package testing or liquid cooling testing methods. Attached Figure Description

[0029] Figure 1 This is a flowchart of Example 1; Figure 2 A structural schematic diagram defining the key structural areas on the cold plate is shown for this embodiment. Figure 3 This is a system block diagram of Example 2; Figure 4 This is a structural diagram illustrating the key areas of the cold plate structure defined in Example 3; Figure 5 This is a schematic diagram showing the logical superposition of the key structural region and the measured thermal characteristic region in Example 3; Figure 6 This is the two-dimensional weight allocation diagram after the values ​​were assigned in Example 3; Figure 7 This is a schematic diagram of the sensor placement in Example 3; Figure 8 The graph shows the fitness and error curves calculated iteratively in Example 3; where (a) is the fitness convergence curve after each iteration; and (b) is the convergence error curve after each iteration. Figure 9 This is a graph showing the convergence results in Example 3; Figure 10 This is a schematic diagram of the structure in Example 3 where 30 sensors are mapped onto the cold plate.

[0030] The diagram is marked as follows: 1. Cold plate, 2. Liquid distribution chamber and cold plate inlet area, 3. Parallel flow channel core heat exchange area, 4. Flow channel fin heat conduction area, 5. Manifold and flow channel outlet area, 6. Geometric symmetry axis area. Detailed Implementation

[0031] Example 1 like Figure 1 As shown, this embodiment provides a method for intelligent sensor placement in a phase change direct cooling system, which includes the following steps: S1. Definition of critical structural regions; Before physical testing, the design drawings and flow channel layout of cold plate 1, as well as historical temperature sensor placement experience data, were analyzed. Based on the mass-heat coupling law of phase change heat transfer, the hydraulic characteristics of fluid flow and the heat conduction mechanism of cold plate 1, the key structural areas that must be monitored on cold plate 1 were predefined.

[0032] The definition of critical structural regions should fully consider areas with uneven refrigerant flow, dryness, complex structural flow, or abrupt changes in heat transfer. These regions are the core control units for the temperature field, phase change heat transfer efficiency, and flow stability of cold plate 1, and their temperature directly determines the overall heat transfer performance, temperature uniformity, and reliability of cold plate 1. Its function is to accurately capture the key physical processes of phase change heat transfer, avoiding missing or misjudged test data.

[0033] like Figure 2 As shown, the key structural region determines the basic shape of the temperature field, which should include the following parts: (1) Liquid distribution chamber and cold plate inlet area 2: This is the upstream starting section of the fluid on the cold plate 1, including the inlet connection section where the fluid just enters the cold plate 1, the pre-buffered liquid distribution chamber, and the branching transition section. This area is the initial state control zone for phase change heat transfer, which determines the start-up of phase change and the uniformity of flow distribution. According to the flow law of branch pipes in fluid mechanics, the channel geometry and inlet velocity in this area will directly lead to the deviation of the working fluid flow rate in each parallel channel. Uneven flow rate will cause insufficient working fluid in some channels and stagnation of working fluid in some channels. Therefore, by placing points in this area, the temperature difference at the inlet of each channel can be monitored, the uniformity of flow distribution can be inferred, the initial temperature of the refrigerant entering the cold plate 1 can be determined, and the subsequent temperature changes can be compared to evaluate the heat transfer efficiency of the refrigerant.

[0034] (2) Parallel flow channel core heat exchange zone 3: This is the dense area of ​​the main flow channel where the fluid inside the cold plate 1 undergoes forced convection heat exchange with the bottom surface of the battery, and it bears the majority of the heat exchange. The fluid continuously completes heat exchange in this area along the predetermined flow path. This area includes core thermal abrupt change areas such as the branch ports inside the cold plate 1, which is the main reaction zone for phase change heat exchange. It directly reflects the heat exchange efficiency and phase change intensity of the cold plate 1. By observing the temperature changes in this area, the core temperature distribution of the entire cold plate 1 can be directly observed. Temperature observation points are set on the parallel flow channels to compare and analyze the rationality of the flow channel design of the cold plate 1, thereby carrying out iterative optimization.

[0035] (3) The heat conduction zone 4 between the flow channels is a solid heat conduction structure between adjacent fluid flow channels on the cold plate 1. This area is a heat compensation and temperature control zone for phase change heat transfer. It can transfer the heat of the heating element to the wall of each flow channel and realize heat compensation between the flow channels at the same time.

[0036] (4) Manifold and flow channel outlet area 5: This is the downstream end section of the cold plate 1, including the confluence section at the end of each branch and the outlet connection section where the fluid flows out of the cold plate 1. This area is the final state control area of ​​phase change heat transfer, reflecting the influence of the degree of phase change completion. By comparing the temperature in this area with the inlet temperature, parameters such as the heat transfer efficiency and superheat of the cold plate 1 can be directly obtained.

[0037] (5) Geometric symmetry axis region 6: This is the solid and flow channel area covered by the axis of symmetry or symmetry plane that satisfies the geometric symmetry (axial symmetry, central symmetry, or mirror symmetry) of the overall shape of the cold plate 1, the flow channel arrangement, and the inlet and outlet layout. This area serves as the reference area for temperature monitoring of the cold plate 1. Under the ideal conditions of geometric symmetry, symmetrical heat flow loading, and symmetrical working fluid distribution of the cold plate 1, the temperature field at the axis of symmetry should be symmetrically distributed. By placing points along this axis, the deviation between the actual temperature field and the ideal symmetry state can be monitored, thereby inferring problems such as uneven flow distribution, symmetrical heat flow loading, and structural processing errors.

[0038] S2. Classification of measured thermal feature regions based on infrared thermal imaging scanning; A battery is placed on the upper surface of the cold plate 1, and the battery is charged and discharged under constant conditions to obtain the typical steady-state operating conditions of the battery. Then, under the typical steady-state operating conditions, an infrared thermal imager is used to perform a full-field scan of the lower surface of the cold plate 1 to obtain a thermal feature spectrum reflecting the actual working state of the cold plate 1. Through thermal feature extraction and classification, the measured thermal feature regions in the thermal feature spectrum are identified, namely the high-temperature region, the low-temperature region, and the high-temperature gradient region.

[0039] The specific process of thermal feature extraction and classification is as follows: Three basic thermal feature parameters are extracted from the thermal feature map: temperature value, temperature difference between adjacent pixels, and regional temperature change rate. First, the upper and lower thresholds of the overall temperature distribution are statistically analyzed across the entire domain. Connected pixel regions with temperatures in the high threshold range (e.g., ≥30℃) are classified as high-temperature regions, and connected pixel regions with temperatures in the low threshold range (e.g., <30℃) are classified as low-temperature regions. Then, a gradient judgment threshold is set based on the magnitude of the change in the absolute value of the temperature gradient between adjacent pixels (e.g., ≥0.5℃ / mm~2℃ / mm). Connected regions with local temperature gradients exceeding the set threshold and drastic temperature changes are classified as high-temperature gradient regions. Finally, the accurate extraction and classification labeling of the three types of measured thermal feature regions are completed.

[0040] By identifying high-temperature zones, low-temperature zones, and high-temperature gradient zones, a set of regions reflecting the key characteristics of the temperature field under actual working conditions of the cold plate 1 can be obtained. These regions reflect the true working state of the cold plate 1 under specific operating conditions. Determining the heat distribution of the cold plate 1 under actual working conditions helps to lay the foundation for integration and comparison with key structural areas.

[0041] It is understandable that the temperature distribution of the upper surface of the cold plate 1 is the same as that of the lower surface. Under typical operating conditions, since the battery is placed on the upper surface of the cold plate 1, the actual thermal characteristic area is obtained by scanning the lower surface of the cold plate 1.

[0042] S3. Integration and optimization based on the key structural region and the measured thermal characteristic region; First, the structural critical regions defined in S1 are projected onto the physical coordinates of the cold plate 1 surface. These regions are areas where uneven flow, drying, complex structural flow, or abrupt changes in heat transfer may occur, based on a pre-judged understanding of the phase change heat transfer mechanism. Then, within the same physical coordinate system, the structural critical regions and the measured thermal characteristic regions are logically superimposed. The superimposed regions include an overlapping area where the structural critical regions and measured thermal characteristic regions overlap, a first non-overlapping area containing only the measured thermal characteristic regions, a second non-overlapping area containing only the structural critical regions, and a third non-overlapping area containing both non-structural critical regions and non-measured thermal characteristic regions. Next, weights are assigned to the superimposed regions using the following method: a first-level weight is assigned to the overlapping areas; second-level, third-level, and fourth-level weights are assigned to the first, second, and third non-overlapping areas, respectively, with the values ​​of the first, second, third, and fourth-level weights decreasing sequentially. This allows us to highlight the common high-temperature portion of the key structural region and the measured thermal characteristic region, and record these regions as the scope of subsequent point placement. This helps to significantly narrow down and clarify the scope and area of ​​point placement, and reduce the amount of calculation required for point placement iteration.

[0043] Specifically, the weighting of the superimposed regions can be shown in the table below:

[0044] After the assignment is completed, a two-dimensional weight distribution map is generated that reflects the combined value of expected and measured information for each location.

[0045] It is understandable that each of the aforementioned locations refers to every physical coordinate point or discretized grid cell on the surface of cold plate 1, and not just the superimposed region. This is because the two-dimensional weight distribution map needs to cover the entire surface of cold plate 1 to provide a basis for global optimization by the genetic algorithm. The superimposed region is only a subset with higher weights, and the other locations also have corresponding basic weights. This ensures that non-critical regions are not completely overlooked during calculation, but the optimization results will naturally favor the high-weight locations.

[0046] Furthermore, the aforementioned expectation refers to the predicted key temperature field regions based on the structure of cold plate 1. Before any actual measurements were conducted, by analyzing the design drawings, flow channel structure, liquid distribution chamber, fins, inlets and outlets of cold plate 1, and combining this with phase change heat transfer and fluid dynamics principles, it can be predicted that high temperature gradients or hot spots are more likely to occur in areas such as the inlet impact zone, flow channel bends, and fin ends. This expected information is embedded in the key structural region of S1. The aforementioned measured information comes from infrared scanning of S2. Therefore, the combined value of the aforementioned expected and measured information is a comprehensive reflection of the a priori value assigned to the key structural region and the measured value assigned to the measured thermal characteristic region.

[0047] Second, place points on the superimposed regions on the two-dimensional weighted distribution map, and adjust the number of points in each region according to the weights. After determining the total number of points, allocate the number or density of points in each region according to the weights, and place points in each region. The coordinates of the points should be integers. The points should be spread out as much as possible to minimize clustering and ensure that the points cover the entire region as much as possible.

[0048] Third, establish the objective function for all points and solve iteratively using a genetic algorithm. After each round of population iteration, call the penalty function to complete the constraint verification. Through selection, crossover, mutation operations and penalty constraint verification during the iteration process, evolve the optimal point coordinate scheme under the given number of sensors.

[0049] Specifically, the iterative solution process using a genetic algorithm is as follows: S3.1, Establishing the computational foundation: Set the number of points to n, establish a coordinate system with any point on cold plate 1 as the origin, determine the coordinates of the n points on cold plate 1, and normalize the coordinates of all points to integers. By normalizing the points to integer coordinates, the position of the points can be quantified, the range of points can be reduced, and the workload of computation can be reduced.

[0050] S3.2, Encoding Process: The binary encoding method of the genetic algorithm is used to convert the coordinates of each point into a binary encoded string. This facilitates the subsequent crossover operation of the genetic algorithm, thereby improving the computational efficiency and convergence of subsequent iterations.

[0051] S3.3, Generating the Initial Population: Given a number of sensors, z initial individuals are randomly generated to form the initial population. Each initial individual corresponds to a set of z placement schemes randomly selected from n preset placement points, and each initial individual corresponds to a unique binary code string, which serves as the initial basis for iteratively selecting the optimal placement points. By assigning initial binary code strings to the z sensors and using randomly generated initial individuals, the universality of the data is ensured, and its sufficiently large range is beneficial to improving accuracy.

[0052] S3.4, Perform iterative operations: Perform multiple rounds of iteration on the initial population. In each round of iteration, the three core operations of selection, crossover, and mutation are completed sequentially. In the selection operation, in each round of iteration, the roulette wheel method is first used to select excellent individuals for crossover. The crossover operation adopts single-point crossover or two-point crossover, which is achieved by randomly exchanging the binary code string fragments of different initial individuals in the initial population, thereby completing the combination optimization of the placement scheme. In the mutation operation, the mutation probability P is set, and the value of P ranges from 0.01 to 0.1. Each mutation is within the preset mutation range C, that is, within the range of the values ​​of all preset placement coordinates of cold plate 1. The binary code string corresponding to the new individual generated after the initial individual is exchanged is randomly flipped to realize the random adjustment of z placement points, gradually optimizing the placement selection results, so as to increase the randomness of the genetic algorithm and improve the reliability of the results.

[0053] This step uses a roulette wheel selection method to retain the best placement schemes, and combines single-point or two-point crossover to achieve the advantageous recombination of placement schemes. Under reasonable low mutation probability and placement coordinate constraints, local placement fine-tuning is completed. This not only accelerates algorithm convergence and escapes local optima, but also ensures the engineering feasibility of the optimization scheme, significantly improving the efficiency and reliability of cold plate 1 placement optimization.

[0054] S3.5, Iterative Selection and Convergence Judgment: In each iteration, the probability I of each individual being selected is calculated using the roulette wheel method. The formula for calculating the probability I of each individual being selected is as follows: Where i = 1, 2, ..., z; In the formula, I The probability of an individual being selected. For the first i The set of z placement points corresponding to each individual; This is the fitness value.

[0055] Then, based on the calculated probability I, z individuals are randomly selected. These z individuals may include duplicate individuals. At the same time, a preset error e and an iteration count of 20 to 50 are set. When the iteration error of the largest fitness value among all individuals is less than the preset error e in each iteration within the set number of iterations, convergence is achieved.

[0056] This step relies on the roulette wheel method to allocate the selection probability according to the fitness ratio of each placement scheme. Based on the fitness, high-quality individuals are retained to accelerate algorithm convergence and prevent the loss of the optimal solution. At the same time, a dual convergence judgment mechanism of iteration number and fitness error is adopted to ensure that the final cold plate 1 placement optimization result has high data stability and calculation accuracy while controlling the amount of computation.

[0057] S3.6, after each iteration, a penalty function is called for verification: The penalty function uses Euclidean distance for verification, setting a minimum distance amin. The distance between each point and any three other points is calculated sequentially, and the minimum distance is taken. The average of the minimum distances of the z points is then calculated as a1. If a1 < amin, the point is considered too dense and is discarded. If the new point after the iteration does not coincide with the set point on the coordinates, the straight-line distance a2 between the new point and the nearest set point is calculated. If a2 < amin, the new point is replaced by the nearest set point. If a2 > amin, the point is deemed unqualified, discarded, and a new point is randomly introduced for the next iteration. This prevents the points from being too dense, which would render the results meaningless and increases the reliability of the data.

[0058] This step uses the Euclidean distance method to construct a penalty function, and combines a preset minimum safety distance to perform dual verification of the density and coordinates of the iterative point layout. Unqualified individuals with excessively dense layouts or severely deviated positions are eliminated and new individuals are added, effectively avoiding the problem of point clustering failure, ensuring the engineering rationality of the point layout plan, and significantly improving the data reliability of the optimization results.

[0059] S3.7, Output Results: After the genetic algorithm converges iteratively, output the optimal point layout coordinate scheme for the z sensors.

[0060] It should be noted that the five structural key regions defined in S1 in this embodiment are based on Figure 2 The design knowledge and historical data of the parallel flow channel phase change direct cooling plate 1 shown are summarized. The typical characteristics of this type of cooling plate 1 are: it has an inlet liquid distribution chamber and an outlet manifold; there are multiple parallel flow channels in the middle; the refrigerant evaporates and absorbs heat in the parallel flow channels; and there are solid fins between the flow channels. This solution is also applicable to other serpentine flow channel cooling plates, spiral flow channel cooling plates, microchannel cooling plates, stamped plate cooling plates, etc., only requiring redefinition of the corresponding structural critical areas. For example, the structural critical areas of a serpentine flow channel cooling plate can be defined as the inlet impact area, the bending and turning area, the outlet area, the middle area of ​​the straight pipe section, and the area near the flow channel wall; the structural critical areas of a microchannel cooling plate can be defined as the inlet distribution area, the microchannel array area, the outlet collection area, and the edge sealing area.

[0061] The solution in this embodiment is the best balance between the structural critical area and the measured thermal characteristic area under the mathematical optimization framework. It integrates the structural critical area and infrared thermographic data, and uses a genetic algorithm to enable a limited number of sensors to accurately cover the key design sites and temperature characteristic areas. This achieves the most comprehensive temperature field characterization with the fewest measurement points, significantly improving testing efficiency and diagnostic accuracy. At the same time, it also ensures that the placement of measurement points is both engineering-specific and individual-adaptable.

[0062] Example 2 like Figure 3 As shown, this embodiment provides a smart sensor deployment system in a phase change direct cooling system, including: The structural critical area definition module is used to define the structural critical area of ​​cold plate 1 based on the design data and historical layout experience data of cold plate 1, and according to the mass-heat coupling law of phase change heat transfer, the hydraulic characteristics of fluid flow and the heat conduction mechanism of cold plate 1 structure. The measured thermal feature area classification module is used to scan the lower surface of the cold plate 1 with an infrared thermal imager under typical steady-state conditions to obtain a thermal feature map reflecting the actual working state of the cold plate 1; and to identify the measured thermal feature areas in the thermal feature map through thermal feature extraction and classification. The fusion and optimization module is used to logically overlay the key structural region with the measured thermal feature region, and assign weights to the overlaid region to generate a two-dimensional weight distribution map that reflects the comprehensive information value of the expected and measured information at each location. It is used to place points on the overlaid region on the two-dimensional weight distribution map and adjust the number of points in each region according to the weights. It is used to establish the objective function for all points, use a genetic algorithm to solve iteratively, and call a penalty function after each round of population iteration to complete the constraint verification. Through selection, crossover, mutation operations and penalty constraint verification in the iteration process, the optimal point coordinate scheme under the given number of sensors is evolved.

[0063] In the fusion and optimization module, the method for logically superimposing the structural critical area and the measured thermal characteristic area is as follows: the structural critical area is projected onto the physical coordinates of the surface of the cold plate 1, and the logical superposition of the structural critical area and the measured thermal characteristic area is achieved in the same physical coordinate system. The superimposed region includes an overlapping area where the structural critical area and the measured thermal characteristic area overlap, a first non-overlapping area containing only the measured thermal characteristic area, a second non-overlapping area containing only the structural critical area, and a third non-overlapping area containing both the non-structural critical area and the non-measured thermal characteristic area.

[0064] The method for assigning weights to the superimposed regions is as follows: assign a first-level weight to the overlapping region, and assign a second-level weight, a third-level weight, and a fourth-level weight to the first non-overlapping region, the second non-overlapping region, and the third non-overlapping region, respectively, with the values ​​of the first-level weight, the second-level weight, the third-level weight, and the fourth-level weight decreasing sequentially.

[0065] The specific process of iteratively solving using a genetic algorithm is as follows: S3.1, Establish the calculation foundation: Set the number of points to n, establish a coordinate system with any point of cold plate 1 as the origin, determine the coordinates of the n points, and normalize the coordinates of all points to integers; S3.2, Encoding Processing: The coordinates of each point are converted into a binary encoded string using a genetic algorithm. S3.3, Generate the initial population: Given the number of sensors z, randomly generate z initial individuals to form the initial population. Each initial individual corresponds to a set of z placement schemes randomly selected from n preset placement points, and each initial individual corresponds to a unique binary code string, which serves as the initial basis for iteratively selecting the optimal placement point. S3.4, Perform iterative operations: Perform multiple rounds of iteration on the initial population. In each round of iteration, the three core operations of selection, crossover, and mutation are completed sequentially. In the selection operation, each round of iteration first uses the roulette wheel method to select excellent individuals for crossover. The crossover operation adopts single-point crossover or two-point crossover, which is achieved by randomly exchanging binary code string fragments of different initial individuals in the initial population, thereby completing the combination optimization of the placement scheme. In the mutation operation, the mutation probability P is set. Each mutation is within the preset mutation range C. The binary code string corresponding to the new individual generated after the initial individual is exchanged is randomly flipped to realize the random adjustment of the z placement points and gradually optimize the placement selection results. S3.5, Iterative Selection and Convergence Judgment: In each iteration, the probability I of each individual being selected is calculated using the roulette wheel method. Based on the calculated probability I, z individuals are randomly selected. At the same time, a preset error e and the number of iterations are set to 20~50. When the iteration error of the largest fitness value among all individuals is less than the preset error e in each iteration within the set number of iterations, convergence is achieved. In S3.5, the probability I of each individual being selected is calculated using the following formula: Where i = 1, 2, ..., z; In the formula, I The probability of an individual being selected. For the first i The set of z placement points corresponding to each individual; This is the fitness value.

[0066] S3.6, After each iteration, a penalty function is called for verification: The penalty function uses the Euclidean distance method for verification. A minimum distance amin is set, and the distance between each placement point and any three other placement points is calculated in turn. The minimum distance is taken, and the average of the minimum distances of z placement points is taken as a1. If a1 < amin, it is judged as too dense, and this individual is eliminated. If the new individual after the iteration does not coincide with the set placement point on the coordinates, the straight-line distance a2 between the placement point and the nearest set placement point is calculated. If a2 < amin, the placement point is replaced by the nearest set placement point. If a2 > amin, it is judged as unqualified, the individual is eliminated, and a new individual is randomly introduced for the next iteration. S3.7, Output Results: After the genetic algorithm converges iteratively, output the optimal point layout coordinate scheme for the z sensors.

[0067] In detail, based on the same innovative concept, each module in the system described in this embodiment corresponds to each step in Embodiment 1. When used, it adopts the same technical means as Embodiment 1 and can produce the same technical effect, which will not be repeated here.

[0068] Example 3 This embodiment experimentally verifies the method of Embodiment 1, as detailed below: Experimental conditions: A certain cold-rolled steel plate has dimensions of 2192mm × 790mm. (e.g., ...) Figure 4 As shown, by analyzing the design drawings and flow channel layout of the cold plate 1, as well as historical layout experience data, the key structural areas are defined on the cold plate 1, such as... Figure 4 The area within the red box. For example... Figure 5 As shown, the measured thermal feature area obtained by thermal feature scanning is then logically superimposed on the defined structural key area in the same physical coordinate system. After superposition, high temperature appears within the pink box. Figure 6 As shown, weights are assigned to the superimposed regions to generate a two-dimensional weight distribution map, where the red area is the overlapping area, the blue area is the first non-overlapping area, the cyan area is the second non-overlapping area, and the yellow area is the third non-overlapping area, to guide the subsequent placement trend and range.

[0069] 1. Set the number of deployment points to n=300 and the number of temperature sensors to z=30, as shown in the table below:

[0070] 2. Select 30 initial individuals, set the maximum number of iterations to 50, convergence error e=0.001, mutation probability P=0.05, use the two-point cross method, and the minimum point spacing amin=50mm.

[0071] Encoding rules: 12 bits in the x-direction, 10 bits in the y-direction, and 22 bits for each point.

[0072] Fitness value: It can be calculated using fitness functions known in the field, and will not be elaborated further.

[0073] 3. The genetic algorithm is used for iterative solution, and the process is as follows: S3.1. Establishing the computational foundation: The origin is selected as the lower left corner of cold plate 1, at (0, 0); 300 candidate points are globally discretized, and the number of candidate points in different regions is allocated according to weights. The coordinates of each point are determined and are integers, serving as the iteration range to facilitate iterative calculations and accelerate computational efficiency, such as... Figure 7 As shown.

[0074] S3.2. Encoding Processing: A certain point (160, 256) in the overlapping area is converted to binary as x=000010100000, y=0100000000, that is, the binary encoding is 0000101000000100000000.

[0075] S3.3. Generate the initial population: Randomly generate 30 initial individuals. Each initial individual is selected from 300 candidate points without repetition. Each point corresponds to a unique binary code string, which constitutes the initial population. For example, the binary code corresponding to the point (160, 256) is 0000101000000100000000.

[0076] S3.4. Iterative Core Operations: Selection: Calculate the fitness values ​​of the initial 30 individuals. Based on the fitness of the whole scheme, the probability of parent generation is allocated, and 30 individuals are repeatedly drawn as the parent generation pool to maintain the total population at 30.

[0077] Crossover: Select two individuals from the initial 30 individuals. For example, parent individual A: 0000101000000100000000, parent individual B: 0001110011011010110011. Randomly generate two cut points: pos1=9, pos2=16, dividing each encoded string into three segments: bits 0-8, bits 9-15, and bits 16-21.

[0078] Segment A: A1[0-8], A2[9-15], A3[16-21]; Segment B: B1[0-8], B2[9-15], B3[16-21].

[0079] Swap the middle section codes to generate two new offspring individual codes: Offspring C = A1 + B2 + A3, Offspring D = B1 + A2 + B3. Similarly, perform crossover on the initial 30 individuals.

[0080] Mutation: Selecting one individual from 30 initial individuals. Example: After crossover, the encoded fragment of the offspring is: 0000101000000100000000. Random numbers between 0 and 1 are generated bit by bit to determine whether to flip the encoding. The mutation probability P = 0.05. If the original 7th bit is 0, and the random number 0.03 < 0.05, a mutation is triggered to change it to 1. All other bits have random numbers greater than 0.05 and remain unchanged. The new code after mutation is: 0000101100000100000000.

[0081] S3.5. Convergence Iterative Process: When the iterative calculation reaches the 43rd generation, the maximum... When the iterative calculation reaches the 44th generation, the maximum The error is 0.0007 < e = 0.001, indicating that the iteration has converged. For example... Figure 8 The fitness and error curves of the iterative calculation shown below illustrate the convergence process; as shown... Figure 9 The convergence result is shown in the graph.

[0082] S3.6. Penalty Function Verification: In the converged population of 30 individuals (30 points), the pairwise Euclidean distance between all individuals is ≥50mm, and there is no clustering of individuals. The penalty term P is then applied. m =0, no need to eliminate or replace individuals.

[0083] S3.7. Iteration result output (unit: mm): The overlapping regions (14) are as follows: Individual 1: (112, 145); Individual 2: (205, 321); Individual 3: (296, 488); Individual 4: (384, 672); Individual 5: (477, 193); Individual 6: (561, 376); Individual 7: (648, 542); Individual 8: (733, 126); Individual 9: (820, 304); Individual 10: (914, 467); Individual 11: (997, 643); Individual 12: (1086, 228); Individual 13: (1172, 405); Individual 14: (1255, 581).

[0084] The first non-overlapping region (9 individuals) are: Individual 15: (1341, 162); Individual 16: (1428, 339); Individual 17: (1510, 507); Individual 18: (1594, 668); Individual 19: (1677, 214); Individual 20: (1762, 391); Individual 21: (1849, 566); Individual 22: (1931, 253); Individual 23: (2015, 429).

[0085] The second non-overlapping region (5 individuals) are: individual 24: (523, 97); individual 25: (786, 274); individual 26: (1042, 441); individual 27: (1316, 188); individual 28: (1579, 363).

[0086] The third non-overlapping region (2 individuals) are: individual 29: (2084, 135); individual 30: (2156, 472).

[0087] like Figure 10 As shown, the temperature sensor placement points are mapped to their relative positions on the cold plate 1 to complete the intelligent placement of the temperature sensors.

[0088] Conclusion: The final iteration results show that the present invention, through intelligent placement method, only requires the design of 30 temperature sensors to achieve temperature coverage and effective acquisition of cold plate 1. In contrast, existing manual placement techniques and methods are estimated to require at least 50 sensors to complete the placement, which greatly saves sensor costs, manual placement, monitoring and data acquisition workload, and has obvious advantages.

[0089] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention. Furthermore, various different embodiments of the present invention can be arbitrarily combined, as long as they do not violate the spirit of the present invention, they should also be considered as the content disclosed by the present invention.

Claims

1. A method for intelligent sensor placement in a phase change direct cooling system, characterized by: Includes the following steps: S1. Definition of critical structural regions; S2. Classification of measured thermal feature regions based on infrared thermal imaging scanning; Under typical steady-state conditions, an infrared thermal imager is used to scan the lower surface of the cold plate (1) to obtain a thermal feature map reflecting the actual working state of the cold plate (1); through thermal feature extraction and classification, the measured thermal feature area in the thermal feature map is identified. S3. Integration and optimization based on the key structural region and the measured thermal characteristic region; The key structural region and the measured thermal characteristic region are logically superimposed, and weights are assigned to the superimposed region to generate a two-dimensional weight distribution map that reflects the comprehensive information value of the expected and measured information at each location. Points are placed in the superimposed regions on the two-dimensional weight distribution map, and the number of points in each region is adjusted according to the weight. The objective function for all sensor placements is established and solved iteratively using a genetic algorithm. After each round of population iteration, a penalty function is called to verify the constraints. Through selection, crossover, mutation operations and penalty constraint verification during the iteration process, the optimal sensor placement coordinate scheme under the given number of sensors is evolved.

2. The intelligent sensor placement method in the phase change direct cooling system according to claim 1, characterized in that: In S3, the specific process of iteratively solving using a genetic algorithm is as follows: S3.1, Establish the calculation basis: Set the number of points to n, establish a coordinate system with any point of the cold plate (1) as the origin, determine the coordinates of the n points, and normalize the coordinates of all points to integers; S3.2, Encoding Processing: The coordinates of each point are converted into a binary encoded string using a genetic algorithm. S3.3, Generate the initial population: Given the number of sensors z, randomly generate z initial individuals to form the initial population. Each initial individual corresponds to a set of z placement schemes randomly selected from n preset placement points, and each initial individual corresponds to a unique binary code string, which serves as the initial basis for iteratively selecting the optimal placement point. S3.4, Perform iterative operations: Perform multiple rounds of iteration on the initial population. In each round of iteration, the three core operations of selection, crossover, and mutation are completed sequentially. In the selection operation, each round of iteration first uses the roulette wheel method to select excellent individuals for crossover. The crossover operation adopts single-point crossover or two-point crossover, which is achieved by randomly exchanging binary code string fragments of different initial individuals in the initial population, thereby completing the combination optimization of the placement scheme. In the mutation operation, the mutation probability P is set. Each mutation is within the preset mutation range C. The binary code string corresponding to the new individual generated after the initial individual is exchanged is randomly flipped to realize the random adjustment of the z placement points and gradually optimize the placement selection results. S3.5, Iterative Selection and Convergence Judgment: In each iteration, the probability I of each individual being selected is calculated using the roulette wheel method. Based on the calculated probability I, z individuals are randomly selected. At the same time, a preset error e and the number of iterations are set to 20~50. When the iteration error of the largest fitness value among all individuals is less than the preset error e in each iteration within the set number of iterations, convergence is achieved. S3.6, After each iteration, a penalty function is called for verification: The penalty function uses the Euclidean distance method for verification. A minimum distance amin is set, and the distance between each placement point and any three other placement points is calculated in turn. The minimum distance is taken, and the average of the minimum distances of z placement points is taken as a1. If a1 < amin, it is judged as too dense, and this individual is eliminated. If the new individual after the iteration does not coincide with the set placement point on the coordinates, the straight-line distance a2 between the placement point and the nearest set placement point is calculated. If a2 < amin, the placement point is replaced by the nearest set placement point. If a2 > amin, it is judged as unqualified, the individual is eliminated, and a new individual is randomly introduced for the next iteration. S3.7, Output Results: After the genetic algorithm converges iteratively, output the optimal point layout coordinate scheme for the z sensors.

3. The intelligent sensor placement method in the phase change direct cooling system according to claim 2, characterized in that: In S3.5, the formula for calculating the probability I of each individual being selected is as follows: Where i = 1, 2, ..., z; In the formula, I The probability of an individual being selected. For the first i The set of z placement points corresponding to each individual; This is the fitness value.

4. The intelligent sensor placement method in the phase change direct cooling system according to claim 1, characterized in that: In S3, the method of logically superimposing the structural key area and the measured thermal characteristic area is as follows: project the structural key area onto the physical coordinates of the surface of the cold plate (1), and realize the logical superposition of the structural key area and the measured thermal characteristic area in the same physical coordinate system.

5. The intelligent sensor placement method in the phase change direct cooling system according to claim 1, characterized in that: In S3, the superimposed region includes an overlapping region where the structural critical region and the measured thermal characteristic region overlap, a first non-overlapping region that includes only the measured thermal characteristic region, a second non-overlapping region that includes only the structural critical region, and a third non-overlapping region that includes both the non-structural critical region and the non-measured thermal characteristic region.

6. The intelligent sensor placement method in the phase change direct cooling system according to claim 5, characterized in that: In S3, the method for assigning weights to the superimposed regions is as follows: assign a first-level weight to the overlapping region, and assign a second-level weight, a third-level weight, and a fourth-level weight to the first non-overlapping region, the second non-overlapping region, and the third non-overlapping region, respectively, with the values ​​of the first-level weight, the second-level weight, the third-level weight, and the fourth-level weight decreasing sequentially.

7. The intelligent sensor placement method in the phase change direct cooling system according to claim 1, characterized in that: In S1, the method for defining the key area of ​​the structure is as follows: analyze the design data and historical layout experience data of the cold plate (1), and define the key area of ​​the structure of the cold plate (1) based on the mass-heat coupling law of phase change heat transfer, the hydraulic characteristics of fluid flow and the heat conduction mechanism of the cold plate (1) structure.

8. The intelligent sensor placement method in the phase change direct cooling system according to claim 7, characterized in that: In S1, the defined key structural areas include the liquid distribution chamber and cold plate inlet area (2), the parallel flow channel core heat exchange area (3), the flow channel inter-fin heat conduction area (4), the manifold and flow channel outlet area (5), and the geometric symmetry axis area (6).

9. The intelligent sensor placement method in the phase change direct cooling system according to claim 1, characterized in that: In S2, a typical steady-state condition refers to: a battery is placed on the upper surface of the cold plate (1), the battery completely covers all flow channel areas in the cold plate (1), the battery is charged and discharged under constant power, the charging and discharging operation time is not less than 1 hour, and after at least 180 seconds of startup, the battery is charged from 0% to 100% or discharged from 100% to 0%.

10. A sensor intelligent placement system in a phase change direct cooling system, characterized in that: include: The structural critical area definition module is used to define the structural critical area of ​​the cold plate (1) based on the design data and historical layout experience data of the cold plate (1), and according to the mass-heat coupling law of phase change heat transfer, the hydraulic characteristics of fluid flow and the heat conduction mechanism of the cold plate (1) structure. The measured thermal feature area classification module is used to scan the lower surface of the cold plate (1) with an infrared thermal imager under typical steady-state conditions to obtain a thermal feature map reflecting the actual working state of the cold plate (1); and to identify the measured thermal feature areas in the thermal feature map through thermal feature extraction and classification. The fusion and optimization module is used to logically overlay the key structural region with the measured thermal feature region, and assign weights to the overlaid region to generate a two-dimensional weight distribution map that reflects the comprehensive information value of the expected and measured information at each location. It is used to place points on the overlaid region on the two-dimensional weight distribution map and adjust the number of points in each region according to the weights. It is used to establish the objective function for all points, use a genetic algorithm to solve iteratively, and call a penalty function after each round of population iteration to complete the constraint verification. Through selection, crossover, mutation operations and penalty constraint verification in the iteration process, the optimal point coordinate scheme under the given number of sensors is evolved.

11. The intelligent sensor deployment system in the phase change direct cooling system according to claim 10, characterized in that: In the fusion and optimization module, the method for logically superimposing the structural key area and the measured thermal feature area is as follows: the structural key area is projected onto the physical coordinates of the surface of the cold plate (1), and the logical superposition of the structural key area and the measured thermal feature area is realized in the same physical coordinate system.

12. The intelligent sensor deployment system in the phase change direct cooling system according to claim 10, characterized in that: In the fusion and optimization module, the superimposed region includes an overlapping region where the structural critical region and the measured thermal characteristic region overlap, a first non-overlapping region that includes only the measured thermal characteristic region, a second non-overlapping region that includes only the structural critical region, and a third non-overlapping region that includes both the non-structural critical region and the non-measured thermal characteristic region.

13. The intelligent sensor deployment system in the phase change direct cooling system according to claim 12, characterized in that: In the fusion and optimization module, the method for assigning weights to the superimposed regions is as follows: a first-level weight is assigned to the overlapping region, and second-level, third-level, and fourth-level weights are assigned to the first, second, and third non-overlapping regions, respectively, with the values ​​of the first-level, second-level, third-level, and fourth-level weights decreasing sequentially.

14. The intelligent sensor deployment system in the phase change direct cooling system according to claim 10, characterized in that: The specific process of iterative solution using a genetic algorithm in the fusion and optimization module is as follows: S3.1, Establish the calculation basis: Set the number of points to n, establish a coordinate system with any point of the cold plate (1) as the origin, determine the coordinates of the n points, and normalize the coordinates of all points to integers; S3.2, Encoding Processing: The coordinates of each point are converted into a binary encoded string using a genetic algorithm. S3.3, Generate the initial population: Given the number of sensors z, randomly generate z initial individuals to form the initial population. Each initial individual corresponds to a set of z placement schemes randomly selected from n preset placement points, and each initial individual corresponds to a unique binary code string, which serves as the initial basis for iteratively selecting the optimal placement point. S3.4, Perform iterative operations: Perform multiple rounds of iteration on the initial population. In each round of iteration, the three core operations of selection, crossover, and mutation are completed sequentially. In the selection operation, each round of iteration first uses the roulette wheel method to select excellent individuals for crossover. The crossover operation adopts single-point crossover or two-point crossover, which is achieved by randomly exchanging binary code string fragments of different initial individuals in the initial population, thereby completing the combination optimization of the placement scheme. In the mutation operation, the mutation probability P is set. Each mutation is within the preset mutation range C. The binary code string corresponding to the new individual generated after the initial individual is exchanged is randomly flipped to realize the random adjustment of the z placement points and gradually optimize the placement selection results. S3.5, Iterative Selection and Convergence Judgment: In each iteration, the probability I of each individual being selected is calculated using the roulette wheel method. Based on the calculated probability I, z individuals are randomly selected. At the same time, a preset error e and the number of iterations are set to 20~50. When the iteration error of the largest fitness value among all individuals is less than the preset error e in each iteration within the set number of iterations, convergence is achieved. S3.6, After each iteration, a penalty function is called for verification: The penalty function uses the Euclidean distance method for verification. A minimum distance amin is set, and the distance between each placement point and any three other placement points is calculated in turn. The minimum distance is taken, and the average of the minimum distances of z placement points is taken as a1. If a1 < amin, it is judged as too dense, and this individual is eliminated. If the new individual after the iteration does not coincide with the set placement point on the coordinates, the straight-line distance a2 between the placement point and the nearest set placement point is calculated. If a2 < amin, the placement point is replaced by the nearest set placement point. If a2 > amin, it is judged as unqualified, the individual is eliminated, and a new individual is randomly introduced for the next iteration. S3.7, Output Results: After the genetic algorithm converges iteratively, output the optimal point layout coordinate scheme for the z sensors.

15. The intelligent sensor deployment system in the phase change direct cooling system according to claim 14, characterized in that: In S3.5, the formula for calculating the probability I of each individual being selected is as follows: Where i = 1, 2, ..., z; In the formula, I The probability of an individual being selected. For the first i The set of z placement points corresponding to each individual; This is the fitness value.

Images