Method and system for temperature control of a composite disc

Abstract

The application discloses a composite disc temperature control method and system, and belongs to the field of semiconductor manufacturing. The method comprises the following steps: dividing the composite disc into multiple independent partitions for real-time sensing, and constructing multiple partition real-time temperature curves; performing temperature difference analysis based on the multiple partition real-time temperature curves, calculating temperature difference data and temperature difference coupling correlation information; dynamically adjusting the multiple independent partitions according to the temperature difference data and the temperature difference coupling correlation information, and obtaining a temperature control result; color coding the composite disc based on the temperature control result, constructing a temperature sensing map for automatic temperature adjustment, generating a temperature adjustment effect for temperature verification, and performing temperature control grading alarm according to verification data. The application solves the technical problem that the product yield is low due to the large difference in material thermal conductivity characteristics of the existing composite disc.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing, and more specifically to a method and system for temperature control of composite disks. Background Technology

[0002] In semiconductor wafer wet cleaning processes, the ceramic-metal composite heating plate is a core component that supports the wafer and provides a stable process temperature. The uniformity of its global temperature field directly determines the consistency of the chemical etching and cleaning reaction rates on the wafer surface, which is a key factor affecting the yield of advanced process semiconductor products. Existing technologies generally adopt a uniform grid partitioning temperature acquisition method, which cannot adapt to the material characteristics of the ceramic layer of the composite plate, which has low thermal conductivity and high corrosion resistance, and the metal layer, which has high thermal conductivity and high mechanical strength. Furthermore, it does not effectively decouple the thermal crosstalk generated between adjacent partitions through heat conduction, resulting in a large deviation in the calculation of the actual temperature of each partition. At the same time, traditional control strategies are difficult to balance temperature control response speed and overshoot suppression, exhibiting defects such as slow convergence at large temperature differences and significant temperature oscillation at small temperature differences, making it difficult to meet the high-precision temperature control requirements of advanced processes for the core process area. Summary of the Invention

[0003] This application provides a composite disc temperature control method and system, aiming to solve the technical problem of low product yield caused by the large differences in the thermal conductivity of existing composite discs, and to achieve the technical effect of...

[0004] In view of the above problems, this application provides a method and system for temperature control of composite discs.

[0005] The first aspect disclosed in this application provides a method for temperature control of a composite disc, the method comprising: The composite disk is divided into multiple independent zones for real-time sensing, and real-time temperature curves for multiple zones are constructed. Temperature difference analysis is performed based on the real-time temperature curves of the multiple zones to calculate temperature difference data and temperature difference coupling correlation information. The multiple independent zones are dynamically adjusted according to the temperature difference data and the temperature difference coupling correlation information to obtain temperature control results. The composite disk is color-coded based on the temperature control results to construct a temperature sensing map for automatic temperature adjustment. The temperature adjustment effect is generated for temperature verification, and a graded temperature control alarm is performed based on the verification data.

[0006] Another aspect of this application discloses a composite disc temperature control system, which includes: The system includes a real-time sensing module for real-time sensing of multiple independent zones within the composite disk, constructing real-time temperature curves for each zone; a temperature difference analysis module for performing temperature difference analysis based on the real-time temperature curves of the multiple zones, calculating temperature difference data and temperature difference coupling correlation information; a dynamic adjustment module for dynamically adjusting the multiple independent zones according to the temperature difference data and the temperature difference coupling correlation information, obtaining temperature control results; and a graded alarm module for color-coding the composite disk based on the temperature control results, constructing a temperature sensing map for automatic temperature adjustment, generating temperature adjustment effect verification, and performing graded temperature control alarms based on the verification data.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: By combining real-time non-uniform partitioning sensing based on material thermal conductivity and historical process temperature field data with thermal crosstalk decoupling technology, the true temperature of each partition is accurately obtained. Based on multi-dimensional temperature difference feature extraction and online recursive identification of directed coupling graph, thermal coupling spatial decoupling is achieved and accurate temperature difference data is calculated. A staged feedforward-feedback composite control strategy is adopted to achieve rapid convergence and steady-state maintenance of the temperature control process. At the same time, a real-time updated temperature sensing map is constructed through multi-level color coding to achieve temperature field visualization and feedforward automatic temperature adjustment. Ultimately, the temperature control accuracy of the composite disk process area is improved, and graded and accurate early warning of temperature control anomalies is achieved, ensuring the process consistency and product yield of semiconductor wafer cleaning.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the composite disk temperature control method is provided for embodiments of this application; Figure 2 A schematic diagram of the composite disc temperature control system is provided for the embodiments of this application.

[0010] Explanation of reference numerals in the attached diagram: Real-time sensing module 11, temperature difference analysis module 12, dynamic adjustment module 13, graded alarm module 14. Detailed Implementation

[0011] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0012] The overall concept of the technical solution provided in this application is as follows: This application provides a composite disk temperature control method and system. It accurately acquires the true temperature of each zone through real-time non-uniform partitioning sensing based on material thermal conductivity characteristics and historical process temperature field data, combined with Kalman filtering thermal crosstalk decoupling technology. A directed coupling graph is constructed based on multi-dimensional temperature difference feature extraction and Granger causality flow analysis, and thermal coupling spatial decoupling is achieved by identifying the thermal conduction coupling coefficient and dynamic response time parameters online. A staged feedforward-feedback composite control strategy is adopted to achieve rapid convergence and steady-state maintenance of the temperature control process. Simultaneously, a real-time updated temperature sensing graph is constructed through multi-level color coding to achieve temperature field visualization and feedforward automatic temperature adjustment. Ultimately, this improves the temperature control accuracy of the composite disk process area, ensuring process consistency and product yield in semiconductor wafer cleaning.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0014] Example 1, as Figure 1 As shown in the embodiment of this application, a composite disc temperature control method is provided, the method comprising: S100: Divide the composite disk into multiple independent partitions for real-time sensing and construct real-time temperature curves for multiple partitions.

[0015] Specifically, the thermal conductivity characteristics of the composite disk material are retrieved first. This involves combining the low thermal conductivity and corrosion resistance of the ceramic phase with the high thermal conductivity and high mechanical strength of the metal phase in the ceramic-metal composite structure. Additionally, the thermal resistance coefficient at the interface between the two layers and historical process temperature field data are collected. This includes historical data on the disk surface temperature distribution under different chemical spray intensities, wafer placement positions, and heating powers during semiconductor cleaning processes. The disk surface area is then non-uniformly divided, specifically into smaller, denser independent sub-regions, where significant differences in thermal conductivity and drastic temperature fluctuations occur during the process, particularly in the wafer center contact area and the key chemical spray rinsing area. The edge support area of ​​the disk, where heat conduction is uniform and temperature changes are gradual, is divided into larger and sparser independent sub-partitions. After generating multiple independent partitions, temperature data is acquired in a synchronous periodic polling acquisition mode. That is, the temperature sensors of all partitions trigger the acquisition action synchronously at a uniform millisecond time interval. The controller then reads the analog voltage signal output by each sensor in the preset partition order and converts it into a digital quantity, forming a raw temperature signal sequence that corresponds one-to-one with the timestamp. This mode can ensure that the temperature data of all partitions are strictly aligned on the time axis, eliminating time deviation for subsequent cross-partition temperature difference comparison and thermal coupling analysis.

[0016] Subsequently, thermal crosstalk decoupling processing was performed on the original temperature signal sequence. Since the composite disk is a continuous thermally conductive medium, heat transfer occurs between adjacent partitions through thermal conduction, i.e., thermal crosstalk. This causes the temperature value collected by a single sensor to simultaneously contain its own true temperature and thermal interference components from adjacent partitions. Therefore, based on a pre-established three-dimensional thermal conduction model of the composite disk, combined with the geometric positional relationship of each partition and the real-time original temperature data, the crosstalk component is removed by using the Kalman filter algorithm to obtain a corrected temperature value that only reflects the thermal state of the partition itself.

[0017] Finally, time-domain analysis is performed based on the corrected temperature values. That is, a temperature time-series dataset for each partition is constructed with time as the horizontal axis and the corrected temperature value as the vertical axis. Equally spaced time-domain discrete points are selected, and an exponential smoothing algorithm is used to predict the temperature change trend for the next 1-2 sampling periods. The historical corrected temperature values ​​and the predicted temperature values ​​are smoothly connected to generate a real-time temperature curve that can continuously reflect the dynamic temperature change process of each partition.

[0018] S200: Perform temperature difference analysis based on the real-time temperature curves of the multiple zones, and calculate temperature difference data and temperature difference coupling correlation information.

[0019] Specifically, the process begins with multidimensional temperature difference feature extraction. Real-time temperature curves of all zones are traversed at millisecond-level sampling intervals synchronized with temperature acquisition. The absolute deviation between the temperature of each zone and the global target temperature is calculated, constructing a continuously changing deviation time series. Next, a sliding time window is introduced—a fixed-length time interval that slides continuously along the time axis with a single sampling step to capture local data segments. In this scheme, the window length is set to 5 sampling periods. Local statistical features such as the mean, variance, peak value, and rate of change of the deviation time series within each window are extracted to form a local feature set. Simultaneously, the spatial temperature difference between any two adjacent zones is calculated, and time-series analysis is performed on these spatial differences to generate a spatial temperature gradient feature set reflecting the rate of change in temperature distribution on the disk surface. Finally, the local feature set containing time-dimensional fluctuation information is correlated and integrated with the spatial temperature gradient feature set containing spatial distribution information to obtain a temperature difference feature set that comprehensively characterizes the non-uniformity of the temperature field.

[0020] Subsequently, real-time heating power data from multiple independent partitions were introduced and aligned one-to-one with the temperature difference feature set according to the same timestamp, forming an aligned dataset containing temperature deviation, spatial gradient, and power input information. This ensures that temperature changes and power adjustments are strictly synchronized on the time axis in subsequent analyses, eliminating analytical errors caused by time misalignment.

[0021] Next, a causal flow analysis is performed. The Granger causality test is used to analyze the statistical causal relationship between the power change of each partition in the aligned dataset and the temperature change of all other partitions, and to identify the causal influence strength parameters of the effect of a power adjustment in one partition on the temperature of other partitions. According to the natural direction of heat conduction, that is, the direction of heat transfer from the high temperature zone to the low temperature zone, these causal influence strength parameters are directionally correlated to construct a directed coupling graph. The nodes in the graph represent each independent partition, the direction of the directed edges represents the actual flow direction of heat conduction, and the weight of the edges represents the strength of the thermal influence between the corresponding partitions.

[0022] Then, the directed coupling graph is traversed for online recursive identification. A recursive least squares algorithm with a forgetting factor is used to update the thermal conduction coupling coefficient between each partition in real time. This coefficient is the ratio of the heat transferred from one partition to the adjacent partition through thermal conduction per unit time to the temperature difference between the two partitions, reflecting the tightness of thermal coupling between the partitions. The coupling coefficient at the interface between the ceramic layer and the metal layer in the composite disk is significantly higher than that inside a single material. At the same time, the temperature response curves of each partition under different power adjustment commands are retrieved from the historical temperature control response database for online recursive identification. The dynamic response time parameter is calculated, which is the time required for the partition temperature to change from the current value to 63.2% of the target value. This reflects how fast the partition responds to changes in heating power. The partition with a higher proportion of metal layer has a shorter response time.

[0023] Subsequently, the calculated thermal conduction coupling coefficient and dynamic response time parameters are structurally integrated to construct temperature difference coupling correlation information containing information on thermal conduction intensity, response speed, and coupling direction. Finally, based on the temperature difference coupling correlation information, spatial decoupling compensation is performed on multiple independent partitions to eliminate the temperature difference calculation deviation caused by mutual thermal interference between partitions, and to obtain the final temperature difference data that can accurately reflect the real temperature difference of each partition.

[0024] S300: Based on the temperature difference data and the temperature difference coupling correlation information, multiple independent zones are dynamically adjusted to obtain temperature control results.

[0025] Specifically, the temperature control state is dynamically divided based on the absolute value of the temperature difference data: when the real-time temperature difference of any zone is greater than 2℃, it enters the rapid approach stage, the core objective of which is to reduce the large deviation to a controllable range in the shortest possible time, and avoid uneven cleaning of semiconductor wafers due to local low temperatures; when the temperature difference is between 0.2℃ and 2℃, it switches to the fine convergence stage, focusing on suppressing adjustment overshoot and ensuring that the temperature steadily approaches the target value; when the temperature difference of all zones is less than 0.2℃, it enters the steady-state maintenance stage, mainly to counteract external interference such as chemical reagent spraying and ambient temperature fluctuations, and maintain the long-term consistency of the overall temperature of the disk.

[0026] Subsequently, based on the divided temperature control stage, and combining the heat conduction direction and heat conduction coupling coefficient of each partition in the directed coupling graph (i.e., the heat transfer rate between adjacent partitions per unit temperature difference), the influence of a temperature change in one partition on other partitions is reflected. Multiple independent partition feedforward compensation quantities are then constructed. Feedforward compensation is a predictive control mechanism that does not rely on temperature deviation feedback. Instead, it calculates in advance the amount of heat that each partition needs to supplement or reduce based on the current temperature difference distribution and the pre-identified heat coupling model, to offset the temperature impact caused by heat conduction in adjacent partitions. For example, if the central partition needs to be rapidly heated by 5°C, according to the directed coupling graph, 65% of its heat will diffuse to the surrounding eight adjacent partitions. Therefore, while calculating the heating amount of the central partition, the feedforward compensation amount of adjacent partitions will be reduced by 15%-25% to avoid temperature overshoot due to passive heat absorption in adjacent partitions.

[0027] Based on this, feedback correction is performed on multiple independent regions using real-time updated temperature difference data to generate feedback adjustment quantities. The feedback correction adopts a piecewise PID algorithm, using a larger proportional coefficient in the rapid approach phase to improve the response speed, reducing the proportional coefficient in the fine convergence phase and introducing an integral term to eliminate steady-state error, and adding a derivative term in the steady-state holding phase to suppress temperature fluctuations. Closed-loop control corrects the feedforward compensation for deviations caused by model errors, material aging, or changes in process parameters.

[0028] Finally, the feedforward compensation and feedback adjustment of each independent partition are algebraically superimposed to generate the final power adjustment command for that partition. The power commands of all partitions are sent to the corresponding independent heating units for execution through the controller with a synchronization period of 100ms. During the execution process, the temperature feedback data of each partition is continuously collected to form an analysis-adjustment-feedback-reanalysis control loop, and finally obtain the temperature control result that meets the process requirements.

[0029] S400: Based on the temperature control results, the composite disc is color-coded, a temperature sensing map is constructed for automatic temperature adjustment, the temperature adjustment effect is generated for temperature verification, and a temperature control graded alarm is performed based on the verification data.

[0030] Specifically, the system is first structured and analyzed into multiple independent partitions. The real-time temperature value of each partition and the preset target temperature difference data bound to the semiconductor cleaning process are extracted. This means that the maximum allowable deviation between each partition and the global target temperature is preset for different process stages. Under the 7nm process, the core process area is ±0.08℃ and the edge auxiliary area is ±0.12℃. The two are established in a one-to-one correspondence through partition ID. Then, the real-time temperature value of each partition is compared with the corresponding preset target temperature difference data point by point to calculate the positive and negative deviation direction parameters, which represent the direction in which the partition temperature is higher or lower than the target value. Positive deviation is overheating and negative deviation is undercooling. The deviation magnitude parameter represents the specific numerical value of the temperature deviation from the target value.

[0031] Based on this, multi-level color coding is carried out to construct a temperature sensing map. First, continuous intervals are divided based on the positive and negative deviation direction parameters and deviation amplitude parameters. For example, the entire deviation range from -0.5℃ to +0.5℃ is evenly divided into 12 non-overlapping sub-intervals. Then, multi-level color matching is performed on all sub-intervals to construct a color mapping lookup table, i.e., a predefined key-value mapping database. The key is the deviation sub-interval, and the value is the corresponding RGB color value. Cool blue tones represent negative deviation, warm red tones represent positive deviation, and pure green represents zero deviation. The larger the deviation amplitude, the higher the color saturation. Then, the deviation parameters of each independent partition are associated with the color. A mapping lookup table is used for mapping and matching to locate the basic color information of each partition. At the same time, the temperature change rate of each partition is calculated, which is the amount of temperature change in the partition within a unit sampling period. Areas with a temperature change rate greater than 0.05℃ / s are marked with bold boundaries to identify multiple color boundary areas. Then, based on the temperature change rates of adjacent partitions, regional temperature convergence analysis is performed to mark transition areas with temperature differences of less than 0.03℃. Finally, a linear gradient fill algorithm is used to smoothly merge the color boundary areas and transition areas to generate a temperature sensing map that can continuously and intuitively display the temperature distribution of the entire disk, which is the data analysis map required by the user.

[0032] After the temperature sensing map is constructed, temperature prediction is performed for three consecutive sampling periods based on the spatiotemporal temperature distribution information it contains, generating a temperature difference trend map. When a temperature deviation exceeding a preset threshold is predicted for a certain zone in the future, a feedforward automatic temperature adjustment command is immediately triggered, i.e., an active adjustment mechanism based on prediction. At the same time, the temperature response sequence during the execution of the temperature adjustment command is collected, i.e., the temperature data of each zone changing over time and the power adjustment sequence, i.e., the actual power output data of each heating unit. Based on these two sets of sequence data, a temperature adjustment effect verification index system including temperature convergence time, steady-state fluctuation amplitude, overshoot, and energy consumption rate is constructed to quantitatively evaluate the effectiveness of this temperature adjustment. If the verification fails, the evaluation results are fed back to the upstream temperature difference analysis stage to dynamically correct the heat conduction coupling coefficient and PID control parameters.

[0033] Finally, based on the temperature control effect verification indicators and real-time temperature difference data, a graded alarm for temperature control is implemented: Level 1 alarm corresponds to a minor anomaly where the temperature difference of a single zone exceeds the preset target value but is less than 0.2℃ and lasts for less than 10 seconds. The system automatically fine-tunes the power and marks the abnormal area with yellow flashing on the temperature sensing graph. Level 2 alarm corresponds to a moderate anomaly where the temperature difference of a single zone exceeds 0.2℃ but is less than 0.5℃, or where three or more zones simultaneously exceed the tolerance. The system issues an audible and visual alarm and automatically records the temperature, power, and process parameters at the time of the anomaly. Level 3 alarm corresponds to a severe anomaly where the temperature difference of any zone exceeds 0.5℃ or the temperature control effect verification indicator fails to meet the standard for three consecutive cycles. The system immediately cuts off the heating power supply to the corresponding area and triggers a process pause procedure to prevent wafer damage and equipment corrosion.

[0034] Furthermore, in the method provided in the application embodiment, the composite disk is divided into multiple independent partitions for real-time sensing to construct real-time temperature curves for multiple partitions. The method includes: retrieving the material thermal conductivity characteristics and historical process temperature field data of the composite disk to non-uniformly divide the disk surface area to generate multiple independent partitions; polling and collecting data from the multiple independent partitions according to a synchronization cycle to obtain an original temperature signal sequence; performing thermal crosstalk decoupling based on the original temperature signal sequence to obtain corrected temperature values ​​for multiple independent partitions; performing time-domain analysis based on the corrected temperature values ​​to determine multiple time-domain discrete points for trend prediction, and constructing real-time temperature curves for the multiple partitions.

[0035] Specifically, the thermal conductivity characteristics of the composite disk material are retrieved first. This includes the low thermal conductivity and high corrosion resistance of the ceramic phase in the ceramic-metal composite structure, the high thermal conductivity and high mechanical strength of the metal phase, as well as comprehensive physical parameters such as the contact thermal resistance at the interface between the two layers and the thermal diffusivity in different regions. Historical process temperature field data is also collected, specifically the cumulative historical records of temperature fluctuation amplitude, frequency of change, and spatial distribution patterns in different regions of the disk surface under different chemical reagent types, spray pressures, wafer placement positions, and heating powers during semiconductor cleaning processes. A density-based spatial clustering algorithm is then used to non-uniformly divide the disk surface. The wafer center contact area and the vertical spraying area for chemical reagents, which have low thermal diffusivity and temperature fluctuations exceeding ±0.3℃ during the process, are divided into high-density independent sub-areas of 5mm×5mm. The disk edge support area and the heating element mounting area, which have high thermal diffusivity and temperature fluctuations less than ±0.1℃, are divided into low-density independent sub-areas of 15mm×15mm. Each sub-area corresponds to an independent platinum resistance temperature sensor and an independent heating unit. While ensuring the temperature acquisition accuracy of key process areas, the total number of sensors is reduced by 30%, effectively reducing hardware costs and subsequent data processing pressure.

[0036] After generating multiple independent partitions, temperature data is acquired using a polling acquisition mode based on a synchronization cycle. The synchronization cycle refers to the time interval between all temperature sensors triggering acquisition actions, set to 100ms to match the temperature response speed of the semiconductor cleaning process. Polling acquisition means that the controller reads the output signal of each sensor sequentially according to the preset partition number order, avoiding bus conflicts caused by multiple sensors transmitting simultaneously. Specifically, at the beginning of each synchronization cycle, the controller sends a synchronization trigger command to all sensors. All sensors simultaneously acquire the current temperature and convert it into an analog voltage signal. Then, the controller reads the analog signal of each sensor sequentially at a rate of 1ms / sensor. The analog signal is converted into a digital quantity by a 16-bit high-precision ADC, and a timestamp consistent with the synchronization trigger time is added to each data point. Finally, a raw temperature signal sequence containing partition ID, timestamp, and raw temperature value is formed.

[0037] Next, thermal crosstalk decoupling processing is performed on the original temperature signal sequence. Since the composite disk is a continuous thermally conductive medium, heat transfer occurs between adjacent partitions through thermal conduction. This results in the original temperature value collected by a single sensor not only containing the true temperature of its own partition but also superimposed with the thermal interference components of adjacent partitions. This phenomenon is called thermal crosstalk, which is particularly significant at the ceramic-metal composite interface, where the crosstalk component can reach up to 15% of the original signal. Based on the three-dimensional thermal conduction model of the composite disk established in advance through finite element analysis, the system constructs a thermal crosstalk matrix by combining the geometric positional relationship of each partition, material thermal resistance parameters, and real-time original temperature data. The Kalman filter algorithm is used to decouple the original temperature signal sequence point by point, eliminating the thermal interference components of adjacent partitions, and obtaining the corrected temperature value that only reflects the thermal state of the corresponding partition itself.

[0038] Finally, time-domain analysis is performed based on the corrected temperature values. That is, with time as the horizontal axis and the corrected temperature value as the vertical axis, a temperature time-series dataset is constructed for each independent partition. Equally spaced discrete points in the time domain are selected, that is, the temperature sampling points corresponding to each synchronization cycle. A quadratic exponential smoothing algorithm that can simultaneously track the temperature horizontal trend and the rate of change is used to predict the temperature change trend for the next 1-2 sampling cycles. Then, the historical corrected temperature values ​​and the predicted temperature values ​​are smoothly connected through cubic spline interpolation to generate real-time temperature curves that can continuously and dynamically reflect the temperature change process of each partition. Each curve is globally updated every 100ms, providing a real-time, continuous and forward-looking data source for subsequent temperature difference analysis.

[0039] Furthermore, in the method provided in the application embodiment, temperature difference analysis is performed based on the real-time temperature curves of the multiple partitions to calculate temperature difference data and temperature difference coupling correlation information. The method includes: extracting multi-dimensional temperature difference features based on the real-time temperature curves of the multiple partitions to generate a temperature difference feature set; aligning the real-time power data of multiple independent partitions with the temperature difference feature set according to timestamps to obtain an aligned dataset; performing causal flow direction analysis based on the aligned dataset to identify causal influence intensity parameters; performing intensity correlation analysis according to the thermal coupling direction based on the causal influence intensity parameters to construct a directed coupling graph; traversing the directed coupling graph for online recursive identification to calculate the thermal conduction coupling coefficient; retrieving historical temperature control response data for online recursive identification to calculate the dynamic response time parameter; integrating the thermal conduction coupling coefficient and the dynamic response time parameter to construct temperature difference coupling correlation information; and performing spatial decoupling compensation on multiple independent partitions based on the temperature difference coupling correlation information to calculate the temperature difference data of the multiple independent partitions.

[0040] Specifically, firstly, multidimensional temperature difference feature extraction is carried out to comprehensively characterize the spatiotemporal non-uniformity of the temperature field: real-time temperature curves of all zones are traversed according to a 100ms sampling interval that is completely synchronized with temperature acquisition, and the absolute deviation between the temperature of each zone and the global process target temperature is calculated point by point to construct a deviation time series that evolves continuously over time; then, a sliding time window is introduced, set as a moving time interval with a length of 5 sampling periods and a step size of 1 sampling period, and statistical features such as the mean, variance, peak value, valley value and first-order rate of change of the deviation time series within each window are extracted to form a local feature set reflecting the fluctuation law of the temperature field in the time dimension; at the same time, the instantaneous spatial temperature difference between any two adjacent zones is calculated, and continuous time series analysis is performed on these spatial differences to generate a spatial temperature gradient feature set reflecting the speed and direction of temperature propagation in space on the disk surface; the local feature set containing dynamic information in the time dimension and the spatial temperature gradient feature set containing spatial distribution information are linked and integrated time-stamped to obtain a temperature difference feature set that can completely characterize the global and local changes of the temperature field.

[0041] Subsequently, real-time heating power data from multiple independent partitions are introduced, which are the actual power values ​​currently output by each independent heating unit. The sampling period is also 100ms. This data is matched one-to-one with the temperature difference feature set according to the same timestamp to form an aligned dataset containing three types of information: temperature deviation, spatial gradient, and power input.

[0042] Next, causal flow analysis is performed based on the aligned dataset. The Granger causality test method is used to statistically analyze the statistical dependence between the power change sequence of each partition and the temperature change sequence of all other partitions. The causal influence strength parameter of the power adjustment of one partition on the temperature of other partitions is identified. The value ranges from 0 to 1. The larger the value, the stronger the explanatory power of the power change of that partition on the temperature change of another partition. In the scheme, the causal influence strength of the central process area on the adjacent edge area is about 0.72, while the influence strength of the edge area on the central area is only 0.28. According to the natural physical direction of heat conduction, that is, the direction of heat transfer from the high temperature area to the low temperature area, these causal influence strength parameters are directionally correlated to construct a directed coupling graph. Each node in the graph represents an independent partition, and the directed edge between the nodes represents the actual flow direction of heat conduction. The weight of the edge corresponds to the causal influence strength parameter, which intuitively presents the complex thermal interaction network between the partitions of the composite disk.

[0043] Then, all directed edges of the directed coupled graph are traversed for online recursive identification. A recursive least squares algorithm with a forgetting factor is used, with the forgetting factor set to 0.98. This approach retains historical data while giving more weight to recent data, automatically adapting to long-term drift factors such as composite disk material aging and changes in interface thermal resistance. The thermal conduction coupling coefficient between adjacent partitions is updated in real time, which is the ratio of the heat transferred between two adjacent partitions per unit time to their temperature difference, reflecting the tightness of thermal coupling between partitions. At the same time, the temperature response curves of each partition under different power step commands are retrieved from the historical temperature control response database and subjected to the same online recursive identification. The dynamic response time parameter, i.e., the time constant of the first-order thermal system, is defined as the time required for the partition temperature to change from the initial value to 63.2% of the target value, reflecting how quickly the partition responds to changes in heating power.

[0044] Subsequently, the calculated thermal conduction coupling coefficient matrix and the dynamic response time parameter vector are structurally integrated to construct temperature difference coupling correlation information containing complete information on thermal conduction intensity, response speed, and coupling direction. Finally, based on the temperature difference coupling correlation information, spatial decoupling compensation is performed on multiple independent partitions. The system constructs a thermal coupling inverse matrix, performs a linear transformation on the original temperature difference vector, removes thermal crosstalk components from adjacent partitions, and obtains the final temperature difference data that accurately reflects the heating requirements of each partition.

[0045] Furthermore, in the method provided in the application embodiment, multidimensional temperature difference feature extraction is performed based on the real-time temperature curves of the multiple partitions to generate a temperature difference feature set. The method includes: synchronously traversing the real-time temperature curves of the multiple partitions according to the sampling interval to calculate the absolute deviation and construct a deviation time series; pre-setting a sliding time window and extracting a local feature set of the deviation time series according to the sliding time window; calculating the difference between adjacent temperatures in the multiple independent partitions to obtain multiple spatial temperature difference values; performing time series analysis based on the multiple spatial temperature difference values ​​to generate a spatial temperature gradient feature set; and associating and integrating the local feature set with the spatial temperature gradient feature set to construct a temperature difference feature set.

[0046] Specifically, firstly, the sampling interval is uniformly set to 100ms, which is completely synchronized with the temperature acquisition and consistent with the power adjustment cycle of the composite disk heating unit. The real-time temperature curves of all independent partitions are synchronously traversed every moment. The absolute deviation between the real-time temperature value of each partition at that moment and the preset global target temperature of the semiconductor cleaning process is calculated. A deviation time series is constructed for each partition, with the timestamp as the index and the elements as the corresponding deviation values ​​at that moment. This series can continuously and orderly reflect the dynamic evolution of the temperature of a single partition from the process target over time. The synchronous traversal design ensures that the deviation data of all partitions are completely aligned on the time axis.

[0047] Next, a sliding time window is preset, using a continuous sliding window with a length of 5 sampling periods (500ms) and a step size of 1 sampling period (100ms). This parameter balances the temporal resolution and noise resistance of feature extraction. The deviation time series of each partition is segmented according to the window, and the mean of the deviation data in each window is extracted to reflect the average deviation level and variance within the window, the degree of fluctuation and peak value of the deviation, the maximum deviation and valley value within the window, the minimum deviation and first-order rate of change within the window, and the statistical characteristics such as the rate of increase or decrease of the deviation. These local statistical characteristics of all partitions are summarized by timestamp to form a local feature set that can comprehensively characterize the dynamic change law of the temperature field in the time dimension. The continuous movement characteristic of the sliding window enables it to completely capture the brief temperature fluctuations caused by instantaneous process operations such as chemical reagent spraying and wafer handling, while fixed window segmentation will miss such key abnormal information due to data fragmentation.

[0048] Subsequently, based on the partitioned topology formed by the previous non-uniform partitioning, the temperature difference was calculated at the same time only for adjacent partitions with a common physical boundary, resulting in multiple spatial temperature difference values. This difference directly reflects the degree of temperature difference between adjacent areas on the disk surface. Since heat conduction mainly occurs in physically contacting adjacent areas, the thermal influence of non-adjacent partitions can be ignored. Therefore, calculating only the adjacent difference values ​​can significantly reduce the amount of computation while ensuring that the physical meaning of the features is clear. Especially at the ceramic-metal composite interface, due to the significant difference in the thermal conductivity coefficients of the two materials, the spatial temperature difference between adjacent partitions is usually 30%-50% higher than that of single-material areas. This feature also provides an important basis for subsequent identification of areas with strong thermal coupling.

[0049] Based on this, a time series analysis is performed on the spatial temperature difference sequence corresponding to each pair of adjacent partitions, that is, the ordered sequence of temperature differences of the adjacent pair at different sampling times. The mean, variance and rate of change of each difference sequence are extracted. Combined with the geometric size of the partition, the spatial temperature difference is converted into the temperature change per unit distance, i.e., the spatial temperature gradient. The gradient time series features of all adjacent partition pairs are summarized by timestamp to generate a spatial temperature gradient feature set that can simultaneously reflect the non-uniformity of the spatial distribution of the temperature field and its evolution over time. Static spatial temperature difference can only describe the temperature distribution at a certain moment, while time series analysis can capture the dynamic process of the temperature gradient caused by heat conduction spreading from the high temperature region to the low temperature region, providing key dynamic information for subsequent causal flow analysis.

[0050] Finally, using the same timestamp, the local temporal features of all partitions at the same moment are spliced ​​and associated with the spatial gradient features of all adjacent pairs point by point to form a high-dimensional feature vector containing complete information in both time and space dimensions at each moment. The feature vectors of all moments together constitute a temperature difference feature set that can comprehensively and accurately characterize the global and local changes of the temperature field of the composite disk.

[0051] Furthermore, in the method provided in the application embodiment, multiple independent zones are dynamically adjusted according to the temperature difference data and the temperature difference coupling correlation information to obtain a temperature control result. The method includes: dividing the temperature control state based on the temperature difference data to generate a rapid approach stage, a fine convergence stage, and a steady-state maintenance stage; performing heating compensation on multiple independent zones according to the directed coupling graph and the thermal conduction coupling coefficient based on the rapid approach stage, the fine convergence stage, and the steady-state maintenance stage to construct feedforward compensation quantities for multiple independent zones; performing feedback correction on multiple independent regions based on the temperature difference data to generate feedback adjustment quantities for multiple independent regions; superimposing the feedforward compensation quantities and feedback adjustment quantities of the multiple independent zones to generate a power adjustment command; and synchronously sending the power adjustment command to multiple independent regions for execution to obtain the temperature control result.

[0052] Specifically, the temperature control state is first divided based on the absolute value of the real-time temperature difference data. That is, according to the control priority and target corresponding to different deviation ranges, the entire temperature control process is dynamically divided into three stages with independent control strategies. When the real-time temperature difference of any partition is greater than 2℃, it enters the rapid approach stage, the goal of which is to reduce the large deviation to a controllable range in the shortest possible time. This is because if the local temperature deviates from the target value for a long time in the semiconductor cleaning process, it will lead to uneven chemical etching rate on the wafer surface and irreversible yield loss. When the temperature difference of all partitions drops to the range of 0.2℃ to 2℃, it automatically switches to the fine convergence stage, focusing on suppressing adjustment overshoot. If high gain control is continued at this time, it is easy to cause large oscillations after the temperature exceeds the target value, prolonging the system stabilization time. When the temperature difference of all partitions is less than 0.2℃, it enters the steady-state maintenance stage, which mainly counteracts the continuous external interference such as chemical reagent spray cooling, ambient temperature fluctuations, and wafer heat exchange, and maintains the long-term consistency of the overall temperature of the disk. Subsequently, based on the divided temperature control stages, and combined with a directed coupling graph—a topological graph that intuitively presents the direction and intensity of heat conduction between each partition—nodes represent independent partitions, directed edges represent the natural direction of heat flow, and the edge weights are the heat conduction coupling coefficient and the heat conduction coupling coefficient, which is the ratio of the heat transferred between two adjacent partitions per unit time to their temperature difference, reflecting the tightness of heat coupling between partitions. Multiple feedforward compensation quantities for independent partitions are then constructed. This feedforward compensation is a model-based predictive control mechanism that does not rely on pre-existing temperature deviation feedback but rather on the current temperature difference distribution and pre-identified heat coupling characteristics. The amount of heat that needs to be added or reduced for each partition is calculated in advance to offset the temperature effect caused by heat conduction from adjacent partitions. For example, if the central process area needs to be heated by 5°C quickly, according to the directed coupling diagram, 65% of its heat will diffuse to the eight adjacent partitions. Therefore, while calculating the heating amount of the central partition, the feedforward compensation amount of the adjacent partitions will be reduced by 15%-25% accordingly. In the rapid approach phase, the feedforward compensation amount accounts for more than 70% of the total control amount to ensure the response speed. In the fine convergence phase, it is reduced to 40% to reduce the impact of model error. In the steady-state holding phase, it is mainly used to compensate for constant heat conduction loss.

[0053] Based on this, feedback correction is performed on multiple independent regions using real-time updated temperature difference data to generate feedback adjustment quantities. Feedback correction is used to correct inherent deviations caused by model approximation, aging of composite disk materials, and changes in process parameters in feedforward compensation. A piecewise PID algorithm is used to implement feedback correction, dynamically adjusting control parameters according to different temperature control stages: a larger proportional coefficient is used in the rapid approach stage to quickly reduce the deviation, and integral and derivative terms are not introduced to avoid system oscillation; the proportional coefficient is reduced in the fine convergence stage, and an integral term is introduced to eliminate steady-state error; in the steady-state holding stage, the proportional coefficient is further reduced and a derivative term is added to suppress small temperature fluctuations.

[0054] Finally, the feedforward compensation and feedback adjustment of each independent partition are algebraically superimposed to generate the final power adjustment command for that partition, in watts, corresponding to the output power of each independent heating unit. The power commands of all partitions are sent to the corresponding independent heating units for execution through the controller with a synchronization period of 100ms. During the execution process, the temperature feedback data of each partition is continuously collected and sent back to the upstream temperature difference analysis stage, forming a control loop of sensing-analysis-adjustment-feedback, and finally obtaining the temperature control result that meets the requirements of the semiconductor cleaning process.

[0055] Furthermore, in the method provided in the application embodiment, the composite disk is color-coded based on the temperature control result, a temperature sensing map is constructed for automatic temperature adjustment, a temperature adjustment effect is generated for temperature verification, and a temperature control graded alarm is performed based on the verification data. The method includes: analyzing the temperature control result according to multiple independent partitions, extracting the real-time temperature values ​​and preset target temperature difference data of multiple independent partitions, wherein the preset target temperature difference data corresponds to the real-time temperature values; comparing the real-time temperature values ​​of the multiple independent partitions with the preset target temperature difference data to determine the positive and negative deviation direction parameters and deviation amplitude parameters; performing multi-level color coding on the multiple independent partitions according to the positive and negative deviation direction parameters and the deviation amplitude parameters to construct a temperature sensing map; performing continuous periodic temperature prediction based on the temperature sensing map to generate a temperature difference change trend map; triggering a feedforward automatic temperature adjustment command based on the temperature difference change trend map, and collecting a temperature response sequence and a power adjustment sequence; performing temperature adjustment analysis based on the temperature response sequence and the power adjustment sequence to construct a temperature adjustment effect verification index; and performing graded early warning based on the temperature adjustment effect verification index to generate multi-level alarm signals for temperature control graded alarm.

[0056] Specifically, the system first performs structured parsing by partition ID to extract the real-time temperature sampling value, sampling timestamp, and heating unit operating status of each independent partition. Simultaneously, it retrieves the preset target temperature difference data bound to the partition from the process parameter database. This data represents the maximum allowable temperature deviation threshold preset according to the different process stages of the semiconductor cleaning process and the functional positioning of the partition. For example, in the 7nm process, the preset target temperature difference for the central process area directly in contact with the wafer is ±0.08℃, while for the edge auxiliary area that only serves a supporting function, it is ±0.12℃. The preset value for each partition establishes a one-to-one correspondence with its real-time temperature value through a unique partition ID, ensuring the accuracy and relevance of the deviation calculation.

[0057] Subsequently, the real-time temperature value of each zone is subtracted from the global process target temperature to obtain the original deviation value with a sign. The sign part is the positive and negative deviation direction parameter. A positive sign indicates that the zone temperature is higher than the target value, i.e., overheating, and a negative sign indicates that it is lower than the target value, i.e., undercooling. The absolute value of the original deviation value is the deviation amplitude parameter, which directly quantifies the severity of the temperature deviation from the process requirements.

[0058] Based on this, multi-level color coding is carried out to construct a temperature sensing map. First, the entire deviation range from -0.5℃ to +0.5℃ is evenly divided into 12 non-overlapping continuous sub-intervals based on the direction parameter. Each sub-interval is assigned a unique RGB color value and pre-stored as a color mapping lookup table, which is a key-value mapping structure. Cool blue tones represent negative deviation, warm red tones represent positive deviation, and pure green represents zero deviation. The larger the deviation, the higher the color saturation, which makes it easier to intuitively judge the degree of abnormality. After matching the deviation parameter of each partition with the lookup table to obtain the basic color, the boundaries of adjacent partitions are then filled with a linear gradient to eliminate color abrupt changes. Finally, a temperature sensing map that is globally updated every 100ms and can continuously display the temperature distribution of the entire disk is generated, which is the data analysis map required by the user.

[0059] After the temperature sensing map is constructed, based on the real-time temperature values, temperature change rates, and thermal coupling relationships of each partition in the map, a quadratic exponential smoothing algorithm is used to perform temperature prediction for three consecutive sampling periods, totaling 300ms. The prediction deviation value of each partition at each future moment is calculated and connected in chronological order to generate a temperature difference trend map reflecting the future evolution trend of the temperature field. When the trend map shows that the prediction deviation of a certain partition will exceed 80% of its preset target temperature difference in the next 1-2 sampling periods, a feedforward automatic temperature adjustment command is immediately triggered. This is an active control command based on prediction, which can fine-tune the heating power 100-200ms in advance to eliminate temperature deviation in its early stages. After the command is issued, the system synchronously collects the temperature response sequence during the temperature adjustment process, the temperature value sequence of each partition and the power adjustment sequence sampled at 100ms intervals, which is the time series of the actual output power of each heating unit.

[0060] Next, based on these two sets of sequences, four temperature regulation effect verification indicators are calculated: temperature convergence time, the time from the issuance of the command to the temperature difference entering the steady state range, maximum overshoot, the maximum extent by which the temperature exceeds the target value during the temperature regulation process, steady-state fluctuation amplitude, the maximum deviation difference over 10 consecutive sampling periods after stabilization, and unit temperature regulation energy consumption. The indicators are compared with the preset process thresholds. If all indicators are met, the temperature regulation is deemed effective. If any indicator is not met, the temperature regulation is deemed unsuccessful. The cause of failure is fed back to the upstream temperature difference analysis link to dynamically correct the heat conduction coupling coefficient and PID control parameters, thereby achieving system self-optimization.

[0061] Finally, based on the real-time deviation amplitude, duration, and temperature control effect verification results, a three-level alarm is implemented: Level 1 alarm corresponds to a minor anomaly where a single zone exceeds the tolerance but is less than 0.2℃ and lasts for less than 10 seconds. The system automatically fine-tunes the power and marks it with a yellow flashing mark on the temperature sensing graph. Level 2 alarm corresponds to a moderate anomaly where a single zone exceeds the tolerance by 0.2℃-0.5℃, three or more zones exceed the tolerance simultaneously, or the temperature control index fails to meet the standard for two consecutive cycles. The system issues an audible and visual alarm and automatically records the abnormal data and pushes it to the maintenance department. Level 3 alarm corresponds to a severe anomaly where any zone exceeds the tolerance by more than 0.5℃ or the temperature control fails and the deviation continues to expand. The system immediately cuts off the heating power supply to the corresponding area, triggers a process pause, and locks the equipment status to prevent wafer scrapping and disk damage.

[0062] Furthermore, in the method provided in the application embodiment, a temperature sensing map is constructed by performing multi-level color encoding on multiple independent partitions according to the positive and negative deviation direction parameters and the deviation amplitude parameters. The method includes: continuously dividing the partitions based on the positive and negative deviation direction parameters and the deviation amplitude parameters to determine multiple sub-intervals; traversing the multiple sub-intervals to perform multi-level color matching and constructing a color mapping lookup table; performing mapping matching and positioning based on the multiple independent partitions and the color mapping lookup table to obtain the basic color information of the multiple independent partitions; traversing the temperature change rate of the multiple independent partitions and combining it with the basic color information of the multiple independent partitions to perform boundary marking and determine multiple color boundary regions; performing regional temperature convergence analysis based on the temperature change rate of the multiple independent partitions and marking multiple transition regions; and performing gradient filling based on the multiple color boundary regions and the multiple transition regions to construct the temperature sensing map.

[0063] Specifically, based on the positive and negative deviation direction parameters of the output, the deviation direction of the zone temperature relative to the global target temperature is characterized. A positive sign represents overheating and a negative sign represents undercooling. This is the basis for distinguishing heating or cooling requirements and the deviation magnitude parameter, which represents the specific numerical value of the temperature deviation from the target value and directly reflects the severity of the anomaly. The entire deviation range is continuously and non-overlappingly divided. In combination with the temperature control accuracy requirements of the semiconductor 7nm and below process cleaning process, the possible temperature deviation range of the composite disk is limited to -0.5℃ to +0.5℃. Exceeding this range will trigger a level three emergency alarm. With 0.08℃ as the minimum resolution interval of the core process area, the entire range is uniformly divided into 12 continuous sub-intervals, of which 6 are negative deviation directions, corresponding to the undercooling state, 6 are positive deviation directions, corresponding to the overheating state, and the zero deviation point is used as a separate dividing point.

[0064] Subsequently, multi-level color matching is performed across all 12 sub-intervals to construct a color mapping lookup table, i.e., a predefined key-value mapping database. The key is each continuous deviation sub-interval, and the value is the corresponding standard RGB color value. A warm and cool color system that conforms to human visual perception is adopted: pure green (RGB:0,255,0) corresponds to the ideal state of zero deviation; cool blue tones, from light blue to dark blue, correspond to negative deviations, and the larger the deviation, the darker the color; warm red tones, from light red to dark red, correspond to positive deviations, and the larger the deviation, the higher the color saturation. At the same time, a unique color index is assigned to each sub-interval to ensure the efficiency of subsequent mapping matching. The multi-level color matching design enables the temperature sensing map to distinguish temperature differences as small as 0.08℃, meeting the visualization needs of advanced processes for minute temperature fluctuations.

[0065] Next, the real-time deviation parameters of each independent partition are mapped and matched with the color mapping lookup table partition by partition. The positive and negative deviation direction and magnitude of the partition are obtained according to the partition ID, the deviation sub-interval to which it belongs is calculated, and the corresponding RGB color value is quickly found through the color index to obtain the basic color information of each partition. For the central process area with a smaller area in the non-uniform division, the corresponding sub-interval division will be further refined to 0.05℃ to ensure higher color resolution in the key areas.

[0066] Based on this, the temperature change rate of each independent zone is calculated, which is the amount of temperature change in the zone within a unit sampling period of 100ms. This reflects the intensity of heat exchange in the area. The larger the value, the faster the heat conduction or heat loss. When the temperature change rate of a zone exceeds the preset threshold of 0.05℃ / s, the outer boundary of the zone is marked with a bold black line and identified as a color boundary area. These areas usually correspond to locations with concentrated heat gradients, such as ceramic-metal composite interfaces and vertical spraying areas of chemical reagents. Boundary marking can help operators quickly identify high-risk areas with abnormally intense heat exchange.

[0067] Subsequently, a regional temperature convergence analysis was performed, comparing the real-time temperature difference between all adjacent zones one by one. When the temperature difference between two adjacent zones was less than 0.03℃, the common boundary between the two zones was marked as a transition region. This analysis can identify regions with continuous and uniform temperature distribution, providing a basis for subsequent smooth filling and avoiding the appearance of harsh color block boundaries that do not match the real temperature field.

[0068] Finally, bilinear gradient filling is performed based on the marked color boundary areas and transition areas. In the color boundary areas, the clear boundaries of the base colors are maintained to highlight the concentrated areas of thermal gradient. In the transition areas, the intermediate colors with continuous transitions are generated by linear interpolation of the RGB components of the base colors of adjacent partitions, so as to achieve smooth fusion between the colors of different partitions. Finally, a temperature sensing map that is globally updated every 100ms and can truly reflect the continuous temperature field distribution of the entire composite disk is generated.

[0069] In summary, the composite disc temperature control method provided in this application has the following technical effects: By combining real-time non-uniform partitioning sensing based on material thermal conductivity and historical process temperature field data with thermal crosstalk decoupling technology, the true temperature of each partition is accurately obtained. Based on multi-dimensional temperature difference feature extraction and online recursive identification of directed coupling graph, thermal coupling spatial decoupling is achieved and accurate temperature difference data is calculated. A staged feedforward-feedback composite control strategy is adopted to achieve rapid convergence and steady-state maintenance of the temperature control process. At the same time, a real-time updated temperature sensing map is constructed through multi-level color coding to achieve temperature field visualization and feedforward automatic temperature adjustment. Ultimately, the temperature control accuracy of the composite disk process area is improved, and graded and accurate early warning of temperature control anomalies is achieved, ensuring the process consistency and product yield of semiconductor wafer cleaning.

[0070] Example 2, based on the same inventive concept as the composite disc temperature control method in the foregoing examples, such as... Figure 2 As shown in the figure, this application embodiment provides a composite disc temperature control system, which includes: The real-time sensing module 11 is used to divide the composite disk into multiple independent zones for real-time sensing and to construct real-time temperature curves for multiple zones; the temperature difference analysis module 12 is used to perform temperature difference analysis based on the real-time temperature curves of the multiple zones, and to calculate temperature difference data and temperature difference coupling correlation information; the dynamic adjustment module 13 is used to dynamically adjust the multiple independent zones according to the temperature difference data and the temperature difference coupling correlation information to obtain temperature control results; the graded alarm module 14 is used to color-code the composite disk based on the temperature control results, construct a temperature sensing map for automatic temperature adjustment, generate temperature adjustment effect for temperature verification, and perform graded temperature control alarms based on the verification data.

[0071] Furthermore, the real-time sensing module 11 is also used to perform the following steps: retrieve the material thermal conductivity characteristics and historical process temperature field data of the composite disk to non-uniformly divide the disk surface area, generating multiple independent partitions; poll and collect data from the multiple independent partitions according to the synchronization cycle to obtain the original temperature signal sequence; perform thermal crosstalk decoupling based on the original temperature signal sequence to obtain the corrected temperature values ​​of the multiple independent partitions; perform time-domain analysis based on the corrected temperature values, determine multiple time-domain discrete points for trend prediction, and construct the real-time temperature curves of the multiple partitions.

[0072] Furthermore, the temperature difference analysis module 12 is also used to perform the following steps: extracting multidimensional temperature difference features based on the real-time temperature curves of the multiple partitions to generate a temperature difference feature set; aligning the real-time power data of multiple independent partitions with the temperature difference feature set according to timestamps to obtain an aligned dataset; performing causal flow direction analysis based on the aligned dataset to identify causal influence intensity parameters; performing intensity correlation analysis according to the thermal coupling direction based on the causal influence intensity parameters to construct a directed coupling graph; traversing the directed coupling graph for online recursive identification to calculate the thermal conduction coupling coefficient; retrieving historical temperature control response data for online recursive identification to calculate the dynamic response time parameter; integrating the thermal conduction coupling coefficient and the dynamic response time parameter to construct temperature difference coupling correlation information; and performing spatial decoupling compensation on multiple independent partitions based on the temperature difference coupling correlation information to calculate the temperature difference data of the multiple independent partitions.

[0073] Furthermore, the temperature difference analysis module 12 is also used to perform the following steps: synchronously traversing the real-time temperature curves of the multiple partitions according to the sampling interval to calculate the absolute deviation and construct a deviation time series; preset a sliding time window and extract the local feature set of the deviation time series according to the sliding time window; calculate the difference between adjacent temperatures in the multiple independent partitions to obtain multiple spatial temperature difference values; perform time series analysis based on the multiple spatial temperature difference values ​​to generate a spatial temperature gradient feature set; and associate and integrate the local feature set with the spatial temperature gradient feature set to construct a temperature difference feature set.

[0074] Furthermore, the dynamic adjustment module 13 is also used to perform the following steps: dividing the temperature control state based on the temperature difference data to generate a rapid approach stage, a fine convergence stage, and a steady-state holding stage; performing heating compensation on multiple independent zones according to the directed coupling graph and the thermal conduction coupling coefficient based on the rapid approach stage, the fine convergence stage, and the steady-state holding stage to construct feedforward compensation quantities for multiple independent zones; performing feedback correction on multiple independent regions based on the temperature difference data to generate feedback adjustment quantities for multiple independent regions; superimposing the feedforward compensation quantities and feedback adjustment quantities of the multiple independent zones to generate a power adjustment command; and synchronously sending the power adjustment command to multiple independent regions for execution to obtain the temperature control result.

[0075] Furthermore, the graded alarm module 14 is also used to perform the following steps: based on the temperature control results, analyze multiple independent zones to extract real-time temperature values ​​and preset target temperature difference data for multiple independent zones, wherein the preset target temperature difference data corresponds to the real-time temperature values; compare the real-time temperature values ​​of the multiple independent zones with the preset target temperature difference data to determine positive and negative deviation direction parameters and deviation amplitude parameters; perform multi-level color coding on the multiple independent zones according to the positive and negative deviation direction parameters and the deviation amplitude parameters to construct a temperature sensing map; perform continuous periodic temperature prediction based on the temperature sensing map to generate a temperature difference change trend map; trigger a feedforward automatic temperature adjustment command based on the temperature difference change trend map to collect a temperature response sequence and a power adjustment sequence; perform temperature adjustment analysis based on the temperature response sequence and the power adjustment sequence to construct a temperature adjustment effect verification index; and perform graded early warning based on the temperature adjustment effect verification index to generate multi-level alarm signals for graded temperature control alarm.

[0076] Furthermore, the graded alarm module 14 is also used to perform the following steps: continuously dividing the data based on the positive and negative deviation direction parameters and the deviation amplitude parameters to determine multiple sub-intervals; traversing the multiple sub-intervals to perform multi-level color matching and construct a color mapping lookup table; performing mapping matching and positioning based on the multiple independent partitions and the color mapping lookup table to obtain the basic color information of the multiple independent partitions; traversing the temperature change rate of the multiple independent partitions and combining it with the basic color information of the multiple independent partitions to perform boundary marking and determine multiple color boundary regions; performing regional temperature convergence analysis based on the temperature change rate of the multiple independent partitions and marking multiple transition regions; and performing gradient filling based on the multiple color boundary regions and the multiple transition regions to construct the temperature sensing map.

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A composite disc temperature control method, characterized in that, The method includes: The composite disk is divided into multiple independent partitions for real-time sensing, and real-time temperature curves of multiple partitions are constructed. Temperature difference analysis is performed based on the real-time temperature curves of the multiple zones to calculate temperature difference data and temperature difference coupling correlation information. Based on the temperature difference data and the temperature difference coupling correlation information, multiple independent zones are dynamically adjusted to obtain temperature control results. Based on the temperature control results, the composite disc is color-coded, a temperature sensing map is constructed for automatic temperature adjustment, the temperature adjustment effect is generated for temperature verification, and a graded temperature control alarm is set according to the verification data.

2. The composite disc temperature control method as described in claim 1, characterized in that, The composite disk is divided into multiple independent partitions for real-time sensing, and real-time temperature curves for multiple partitions are constructed. The method includes: The material thermal conductivity characteristics and historical process temperature field data of the composite disk are retrieved to divide the disk surface area non-uniformly, generating multiple independent partitions. The original temperature signal sequence is obtained by polling and collecting data from multiple independent partitions according to the synchronization period. Thermal crosstalk decoupling is performed based on the original temperature signal sequence to obtain corrected temperature values ​​for multiple independent partitions; Based on the corrected temperature value, time-domain analysis is performed to determine multiple time-domain discrete points for trend prediction, and real-time temperature curves for the multiple zones are constructed.

3. The composite disc temperature control method as described in claim 1, characterized in that, Based on the real-time temperature curves of the multiple zones, temperature difference analysis is performed to calculate temperature difference data and temperature difference coupling correlation information. The method includes: Multidimensional temperature difference features are extracted based on the real-time temperature curves of the multiple zones to generate a temperature difference feature set. Real-time power data from multiple independent partitions are aligned with the temperature difference feature set by timestamp to obtain an aligned dataset; Based on the aligned dataset, causal flow analysis is performed to identify causal influence strength parameters; Intensity correlation analysis is performed according to the thermal coupling direction based on the aforementioned causal influence intensity parameters to construct a directed coupling graph; The directed coupling graph is traversed for online recursive identification, the thermal conduction coupling coefficient is calculated, historical temperature control response data is retrieved for online recursive identification, and the dynamic response time parameter is calculated. The thermal conduction coupling coefficient is integrated with the dynamic response time parameter to construct temperature difference coupling correlation information; Based on the temperature difference coupling correlation information, spatial decoupling compensation is performed on multiple independent partitions, and the temperature difference data of the multiple independent partitions are calculated.

4. The composite disc temperature control method as described in claim 3, characterized in that, Based on the real-time temperature curves of the multiple zones, multi-dimensional temperature difference features are extracted to generate a temperature difference feature set. The method includes: The absolute deviation is calculated by synchronously traversing the real-time temperature curves of the multiple partitions according to the sampling interval, and a deviation time series is constructed. A preset sliding time window is used to extract the local feature set of the deviation time series according to the preset sliding time window; Based on the multiple independent partitions, the temperature difference between adjacent regions is calculated to obtain multiple spatial temperature difference values; A time series analysis is performed based on the multiple spatial temperature differences to generate a spatial temperature gradient feature set. The local feature set is associated and integrated with the spatial temperature gradient feature set to construct a temperature difference feature set.

5. The composite disc temperature control method as described in claim 3, characterized in that, The method involves dynamically adjusting multiple independent zones based on the temperature difference data and the temperature difference coupling correlation information to obtain temperature control results. Based on the temperature difference data, the temperature control state is divided into three stages: rapid approach stage, fine convergence stage, and steady-state maintenance stage. Based on the rapid approach phase, the fine convergence phase, and the steady-state maintenance phase, heating compensation is performed on multiple independent partitions according to the directed coupling graph and the thermal conduction coupling coefficient to construct feedforward compensation quantities for multiple independent partitions; Based on the temperature difference data, feedback correction is performed on multiple independent regions to generate feedback adjustment amounts for multiple independent regions. The feedforward compensation amount and the feedback adjustment amount of the multiple independent partitions are superimposed to generate a power adjustment command; The power adjustment command is simultaneously sent to multiple independent regions for execution to obtain the temperature control result.

6. The composite disc temperature control method as described in claim 1, characterized in that, Based on the temperature control results, the composite disk is color-coded, a temperature sensing map is constructed for automatic temperature adjustment, the temperature adjustment effect is generated for temperature verification, and a graded temperature control alarm is performed based on the verification data. The method includes: Based on the temperature control results, the data is analyzed according to multiple independent partitions to extract the real-time temperature values ​​and preset target temperature difference data of multiple independent partitions. The preset target temperature difference data corresponds to the real-time temperature values. The positive and negative deviation direction parameters and deviation amplitude parameters are determined by comparing the real-time temperature values ​​of the multiple independent zones with the preset target temperature difference data. A temperature sensing map is constructed by performing multi-level color coding on multiple independent partitions according to the positive and negative deviation direction parameters and the deviation amplitude parameters. Based on the temperature sensing map, continuous periodic temperature prediction is performed to generate a temperature difference change trend map. Based on the temperature difference change trend graph, a feedforward automatic temperature adjustment command is triggered, and the temperature response sequence and power adjustment sequence are collected. Temperature regulation analysis is performed based on the temperature response sequence and the power adjustment sequence to construct a temperature regulation effect verification index. Based on the temperature control effect verification indicators, graded early warnings are generated, and multi-level alarm signals are used for graded temperature control alarms.

7. The composite disc temperature control method as described in claim 6, characterized in that, The method involves performing multi-level color coding on multiple independent zones according to the positive and negative deviation direction parameters and the deviation amplitude parameters to construct a temperature sensing map, including: Based on the positive and negative deviation direction parameters and the deviation amplitude parameters, multiple sub-intervals are continuously divided to determine the following; The multiple sub-intervals are traversed to perform multi-level color matching, and a color mapping lookup table is constructed. Based on the mapping and matching of the multiple independent partitions with the color mapping lookup table, the basic color information of the multiple independent partitions is obtained. By iterating through the temperature change rate of the multiple independent partitions and combining the basic color information of the multiple independent partitions, boundary marking is performed to determine multiple color boundary regions. Based on the temperature change rate of the multiple independent zones, regional temperature convergence analysis is performed, and multiple transition regions are marked. The temperature sensing map is constructed by gradient filling based on the multiple color boundary regions and the multiple transition regions.

8. A composite disc temperature control system, characterized in that, For implementing the composite disc temperature control method according to any one of claims 1 to 7, the system comprises: The real-time sensing module is used to divide the composite disk into multiple independent partitions for real-time sensing and to construct real-time temperature curves for multiple partitions. The temperature difference analysis module is used to perform temperature difference analysis based on the real-time temperature curves of the multiple zones, and to calculate temperature difference data and temperature difference coupling correlation information. The dynamic adjustment module is used to dynamically adjust multiple independent zones based on the temperature difference data and the temperature difference coupling correlation information to obtain temperature control results; The graded alarm module is used to color-code the composite plate based on the temperature control results, construct a temperature sensing map for automatic temperature adjustment, generate temperature adjustment effect for temperature verification, and perform graded temperature control alarms based on the verification data.

Images