Multi-target parameter optimization method, device, system and product of nuclear radiation sensor

Abstract

The invention discloses a multi-target parameter optimization method, device, system and product of a nuclear radiation sensor, relates to the field of nuclear radiation detection, and aims to improve the environmental universality of the sensor and realize the multi-target parameter collaborative optimization design of the nuclear radiation sensor. According to the invention, parameter calibration suitable for various scenes is carried out on the sensor; the preset performance index of the calibrated sensor is used as a constraint condition, cost minimization and / or sales volume maximization are / is used as a target, technological parameters and product parameters of the sensor are optimized, and the performance index, the cost and the sales volume are reasoned through corresponding prediction models. According to the invention, the universality and reliability of the sensor in a complex environment are improved, intelligent and collaborative optimization of multi-target parameters is realized, and the research and development cost and period are reduced.

Description

Technical Field

[0001] This invention relates to the field of nuclear radiation detection technology, and in particular to a method, apparatus, system and product for optimizing multi-target parameters of a nuclear radiation sensor. Background Technology

[0002] With the widespread application of nuclear technology in environmental monitoring, food safety, and medical protection, the performance requirements for nuclear radiation sensors are increasing. Traditional sensor designs rely heavily on physical models and engineering experience, lacking systematic quantitative analysis of the impact of environmental parameters under various practical application scenarios. This makes it difficult to optimize nuclear radiation sensors in terms of reliability, stability, cost, and market acceptance, thus hindering the technological development of nuclear radiation sensor design.

[0003] Known design methods for nuclear radiation sensors primarily rely on physical mechanism models and extensive trial-and-error experiments, which have the following significant shortcomings: First, there is insufficient quantitative analysis of the coupling effects of multiple environmental parameters such as temperature, humidity, air pressure, and background radiation, making it difficult to accurately predict the sensor's performance in complex real-world application scenarios (such as high-humidity marine environments and medical radiation environments). Second, sensor parameter design often focuses on a single performance indicator, lacking a systematic optimization method among multiple objectives such as performance, cost, and market acceptance, resulting in long design cycles, high costs, and weak scenario adaptability. Specifically, existing sensor simulation software (such as physical simulation tools based on finite element analysis) mainly relies on precise physical equations, making it difficult to quickly assess the comprehensive impact of small changes in process parameters on cost and market sales. While general data analysis platforms (such as SPSS and Python sklearn) can perform regression analysis and model training, they lack a dedicated optimization framework for the multi-parameter coupling and high reliability constraints of nuclear radiation sensors, failing to achieve a complete closed loop from performance verification to parameter optimization, and can only obtain the more preferred feasible solution through extensive trial-and-error experiments. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, system, and product for optimizing multi-objective parameters of nuclear radiation sensors, addressing all or part of the aforementioned problems, so as to achieve collaborative optimization design of multi-objective parameters of nuclear radiation sensors while meeting the performance requirements of different scenarios.

[0005] The technical solution adopted in this invention is as follows: In a first aspect, the present invention provides a method for optimizing multiple target parameters of a nuclear radiation sensor, comprising: Perform parameter calibration on the sensor to adapt to various scenarios; Using the predetermined performance indicators of the calibrated sensor as constraints, and with the goal of minimizing cost and / or maximizing sales volume, the process parameters and product parameters of the sensor are optimized; wherein, the performance indicators are determined by the process parameters and environmental parameters of the sensor; the cost is determined by the process parameters and product parameters of the sensor; and the sales volume is determined by the product parameters and price parameters of the sensor.

[0006] In a second aspect, the present invention provides a multi-target parameter optimization device for a nuclear radiation sensor, comprising a processor and a storage medium, wherein the storage medium stores computer instructions, and the processor executes the computer instructions to perform the aforementioned multi-target parameter optimization method for a nuclear radiation sensor.

[0007] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the above-described method for optimizing multi-target parameters of a nuclear radiation sensor.

[0008] In a fourth aspect, the present invention provides a computer program product comprising a computer program that, when executed by a processor, performs the aforementioned method for optimizing multi-objective parameters of a nuclear radiation sensor.

[0009] In a fifth aspect, the present invention provides a multi-objective parameter optimization system for a nuclear radiation sensor, comprising: The user interaction module is used to receive external data and pass it to the processing module, or to output data output by the processing module to the outside. The processing module is used to execute the above-described method for optimizing multi-target parameters of nuclear radiation sensors.

[0010] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. Improve the performance reliability and applicability of nuclear radiation sensors in various complex real-world environments: By establishing the correlation between environmental parameters and performance indicators to calibrate sensor parameters, it is possible to accurately predict the performance of sensors in various real-world scenarios and ensure that they meet the assessment requirements in diverse environments.

[0011] 2. Achieve data-driven and automated optimization of sensor design: Utilize machine learning models to establish mapping relationships between process parameters, product parameters, performance, cost, and sales volume. Combine this with optimization algorithms to achieve intelligent optimization of multi-objective parameters, significantly reducing R&D costs and timelines.

[0012] 3. Significantly reduce manufacturing costs or increase market acceptance: While meeting performance indicators, manufacturing costs can be effectively reduced by synergistically optimizing process and product parameters. At the same time, sales forecasting models can guide product pricing and market strategies, thereby enhancing product market competitiveness.

[0013] 4. Form an intelligent closed-loop iterative system: Establish a complete closed loop of "data acquisition - model training - parameter optimization - experimental verification - feedback update" to achieve continuous improvement and iterative optimization of sensor design.

[0014] 5. Enhance adaptability to multiple scenarios: Through comprehensive analysis of data from different scenarios (marine environment, food safety testing, medical protection, etc.), the sensor design can take into account the special needs of multiple application scenarios, thereby improving the versatility and adaptability of the product.

[0015] 6. Provide an intuitive and easy-to-use interactive interface: Visualize the operation of methods through the web system, reduce the threshold for use, improve the work efficiency of engineers and decision-makers, and support remote collaboration and multi-scenario deployment.

[0016] 7. Achieve cross-domain knowledge integration and collaborative optimization: By using data and models to uniformly represent and collaboratively optimize knowledge from multiple fields such as nuclear detection physics, manufacturing processes, cost economics, and marketing, the design limitations of traditional single disciplines are broken through, multi-objective conflict problems are solved, and the overall optimal solution is obtained. Attached Figure Description

[0017] The present invention will be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of the operation of the multi-objective parameter optimization method for nuclear radiation sensors.

[0018] Figure 2 This is a flowchart of the parameter calibration process for nuclear radiation sensors.

[0019] Figure 3 This is a data flow diagram of a multi-objective parameter optimization method for nuclear radiation sensors in one embodiment.

[0020] Figure 4 This is a diagram of the operating interface of a multi-target parameter optimization system for nuclear radiation sensors in one embodiment. Detailed Implementation

[0021] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0022] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0023] Most known nuclear radiation sensor design methods involve extensive trial-and-error experiments using physical mechanism models constructed in a laboratory environment to determine the optimal target parameters. This approach limits the sensor performance testing and verification to a single scenario, neglecting the coupling relationships between the sensor and different scenarios and environmental parameters (such as temperature, humidity, air pressure, and background radiation intensity). It fails to accurately measure the sensor's performance in various environments. Furthermore, trial-and-error-based sensor parameter optimization methods lack accurate test data in different scenarios, struggle to address multi-objective optimization goals, and require extensive experimentation to obtain optimal design parameters, resulting in excessively long design cycles and low efficiency.

[0024] To address the shortcomings of known nuclear radiation sensor design methods, such as long design cycles, weak scenario adaptability, poor reliability of design parameters, and inability to accommodate multiple design parameters, this application provides a multi-objective parameter optimization method for nuclear radiation sensors. The aim is to achieve collaborative optimization design of multiple objective parameters of nuclear radiation sensors to meet predetermined performance requirements.

[0025] like Figure 1 As shown in the embodiments of this application, the multi-objective parameter optimization method for a nuclear radiation sensor (hereinafter referred to as the sensor) includes the following steps: S1. Perform parameter calibration on the sensor to adapt to various scenarios.

[0026] Experiments have revealed that sensor performance indicators exhibit varying coupling relationships with different complex real-world application scenarios (such as high-humidity marine environments, food safety testing, and medical radiation environments) and different environmental parameters (such as temperature, humidity, air pressure, and background radiation intensity). In this embodiment, the sensor's performance indicators are calibrated to accurately couple with combinations of different application scenarios and environmental parameters. This enables reliable prediction of performance indicators for different environmental parameters under various scenarios, providing reliable driving data for optimizing sensor target parameters. While ensuring scenario adaptability, this also improves the reliability of target parameter optimization.

[0027] In one alternative implementation, such as Figure 2 As shown, step S1 includes the following sub-steps: S11. Under the action of a standard radiation source, collect test data on the performance indicators of the sensor under various scenarios and different environmental parameters.

[0028] For example, in laboratories and various real-world scenarios (such as ports, hospitals, and food processing plants), standard radiation sources are used to act on sensors to record different environmental parameters (such as temperature T, humidity M, air pressure P, and background radiation intensity R) and their corresponding performance indicators (such as energy resolution, energy detection range, sensitivity, and detection efficiency) to obtain test data.

[0029] S12. Preprocess the test data.

[0030] Preprocessing of test data typically includes data cleaning, outlier removal, and data normalization.

[0031] S13. Establish the correlation between sensor performance indicators and usage scenarios and environmental parameters.

[0032] Under the influence of a standard radiation source with known radiation intensity and energy spectrum, a large number of performance index test data are sampled under different usage scenarios and environmental parameters. Combined with data recorded in laboratory scenarios as standard reference values, multiple measurements are performed under controlled variables to compare the mean, variance, etc., of the test data in various scenarios. A regression model or machine learning model is used to establish a mapping relationship (linear or nonlinear) between sensor performance indicators and usage scenarios and environmental parameters, ensuring that the test data in various scenarios are as close as possible to the standard reference values. Based on the fitted or trained model, it is possible to extend the inference to scenarios and environmental parameters that may be involved in actual applications but not covered by the samples. This allows for the inference of sensor performance indicators under different combinations of scenarios and environmental parameters, ensuring that the sensor can meet the performance accuracy standards in various real-world scenarios. This ensures the sensor's applicability to different scenarios, enabling reliable measurements in various usage scenarios, and providing accurate and reliable driving data for subsequent optimization of target parameters.

[0033] Traditional sensor design methods often rely on physical mechanism models to calculate performance indicators by applying different environmental parameters. However, these physical mechanism models neglect the coupling relationship between the sensor and the environment in different scenarios, leading to discrepancies between the calculated performance indicators and actual values ​​when using a uniform physical mechanism model. In this embodiment, however, sensor parameter calibration in step S1 minimizes or eliminates errors that may arise from scene changes in sensor performance testing, thus eliminating uncertainty in performance testing and providing reliable driving data for target parameter optimization.

[0034] Inference about sensor-related parameters under different scenarios and parameters can employ various regression models, such as multivariate statistical regression models, or machine learning models that consider nonlinear mappings, such as neural networks, random forests, and XGBoost, to train and obtain the correlation between output and input parameters. Parameter inference through model training can significantly improve the efficiency of data measurement and optimization compared to trial-and-error experiments.

[0035] The general expression for defining the model is: ; in, The input variables include all environmental parameters, process parameters, etc., and may contain one or more dimensions; These are the model parameters, i.e., the parameters that need to be learned from the training data; Represents the model operation function; This represents the model's output variables, which may include one or more dimensions.

[0036] For example, when fitting / training the relationship between environmental parameters and performance indicators, the output variables of the model are defined. For performance indicators For example, energy resolution, energy detection range, sensitivity, detection efficiency, etc.; input variables The environmental parameters included are: temperature T, pressure P, air humidity M, and background radiation intensity R; corresponding to each input variable component, the model parameters... include Model operation function Represented as: .

[0037] For each performance metric, a corresponding inference model is fitted / trained based on the collected sample data (i.e., the recorded test data) (i.e., the corresponding model parameters are obtained). (See appendix) Figure 3 .

[0038] Considering that environmental parameters and sensor performance indicators are often not linearly related and may have cross-effects, in order to obtain better prediction results, in one optional implementation of this application, a regression model is constructed using a nonlinear model based on the amount of data collected, while a linear model is used for statistical analysis of the effects of related factors.

[0039] S2. Using the predetermined performance indicators of the calibrated sensor as constraints, and with the goal of minimizing costs and / or maximizing sales, optimize the sensor's process parameters and product parameters.

[0040] Step S1 calibrates the sensor parameters, making it suitable for optimizing target parameters in various scenarios, rather than being applicable to only a single scenario as is the case with current sensor design methods.

[0041] Step S2 involves several relationships related to usage conditions or optimization objectives, specifically including: (1) Performance indicators.

[0042] Performance metrics are determined by the sensor's process parameters and environmental parameters. In other words, sensor performance metrics are inferred based on the sensor's process parameters and environmental parameters (the environmental parameters of the current operating environment).

[0043] As an optional implementation method, the reasoning method for performance metrics includes: S21. Using the trained performance prediction model, inference is performed based on the current sensor process parameters and environmental parameters to obtain the sensor's performance indicators.

[0044] The performance prediction model is obtained by training a machine learning model based on different process parameters, environmental parameters and corresponding performance indicators of historically collected sensors.

[0045] Different types of radiation particles may require sensors with different specifications, processes, or principles, and different performance evaluation indicators. Specifications, processes, and other information, like environmental parameters, determine the sensor's performance indicators. Establishing a performance prediction model helps optimize relevant parameters such as processes and specifications according to the requirements of the evaluation indicators, thereby achieving the expected performance that meets the evaluation requirements.

[0046] The performance prediction model can still be represented by the general expression mentioned above. However, in practical applications, the output variables of the performance prediction model... For performance indicators It is represented by multiple components (such as energy resolution, energy detection range, sensitivity, detection efficiency, etc.); input variables Including process parameters and environmental parameters (Same as above, including, for example, temperature T, air pressure P, air humidity M, and background radiation intensity R); Model parameters to be trained Represented as Model operation function Represented as Then we have: .

[0047] Different process parameters based on historically acquired sensors Environmental parameters and corresponding performance indicators Optimize training model parameters Thus, a performance prediction model is obtained.

[0048] Performance prediction model parameters The optimization process includes: First, collecting data containing process parameters Environmental parameters and corresponding performance indicators The historical dataset was used and divided into training and testing sets; secondly, methods such as grid search or Bayesian optimization were employed to optimize the model parameters. Fine-tuning was performed; finally, the model was trained using the training set, and its prediction accuracy (such as mean squared error and coefficient of determination) was validated using the test set. (and other indicators).

[0049] In one alternative implementation, the sensor's process parameters Three aspects were considered: 1) Sensor component fabrication process parameters. This includes processing, packaging, and integration process parameters based on materials such as CZT and PIPS, such as the cutting precision of CZT material and the ion implantation dose of PIPS.

[0050] 2) Circuit design and fabrication process parameters. This covers the design and fabrication processes of low-noise readout circuits, data processing modules, power supply and communication modules, such as wiring spacing and soldering precision.

[0051] 3) Batch production process parameters. These include process parameters related to large-scale production, such as component quality control and production line process control, for example, assembly tolerances and cleanliness control rates.

[0052] Through the above structured data and systematic modeling process, a performance prediction model that can accurately reflect the complex mapping relationship between process parameters, environmental parameters and performance indicators is constructed.

[0053] For example, for a performance prediction model, its input variables include: Process parameters: [Detector size, bias voltage, ion implantation metering, circuit wiring spacing]; Environmental parameters: [T, P, M, R]; Output variables for: Performance metrics: [Energy resolution, sensitivity].

[0054] (2) Cost.

[0055] Cost is determined by the sensor's process parameters and product parameters. In other words, the sensor cost is predicted based on its current process and product parameters.

[0056] Similar to the reasoning methods for performance metrics, in one alternative implementation, the reasoning method for cost includes: S22. Using the trained cost prediction model, inference is made based on the current sensor process parameters and product parameters to obtain the sensor cost.

[0057] The cost prediction model is trained on a machine learning model based on different process parameters, product parameters, and corresponding costs collected from historical sensors.

[0058] The construction of the cost prediction model is similar to that of the performance prediction model; it can also be represented by the general expression mentioned earlier. The output variables of the cost prediction model... For cost Input variables Including process parameters (Same as above) and product parameters (e.g., basic performance, casing material, interface type, display screen size, etc.); parameters of the model to be trained. Represented as Model operation function Represented as Then we have: .

[0059] Different process parameters based on historically acquired sensors Product parameters and corresponding costs Optimize training model parameters Thus, the cost prediction model is obtained. The optimization process of the cost prediction model is similar to that of the performance prediction model, and its input and output examples are also similar to those of the performance prediction model, so no separate examples will be given here.

[0060] (3) Sales volume.

[0061] Sales volume is determined by the sensor's product parameters and price parameters. In other words, sensor sales volume is predicted based on the sensor's current product parameters and price parameters.

[0062] Similar to the reasoning methods for performance metrics, in one alternative implementation, the reasoning method for sales volume includes: S23. Using the trained sales forecasting model, infer the sales volume of the sensor based on the current product parameters and price parameters of the sensor.

[0063] The sales forecasting model is trained on a machine learning model based on different product parameters, price parameters, and corresponding sales volumes collected from historical sensors.

[0064] The construction of the sales forecasting model is similar to that of the performance forecasting model; it can also be represented by the general expression mentioned earlier. The output variables of the cost forecasting model... For sales Input variables Including product parameters (Same as above) and price parameters Model parameters to be trained Represented as Model operation function Represented as Then we have: .

[0065] Different product parameters of sensors based on historical data. Price parameters and corresponding sales Optimize training model parameters Thus, a sales forecasting model is obtained. The optimization process of the sales forecasting model is similar to that of the performance forecasting model, and its input and output examples are also similar to those of the performance forecasting model, so no separate examples will be given here.

[0066] In general, based on historically collected environmental parameters, process parameters, product parameters, cost, price parameters, and sales volume of the sensor under various scenarios, parameter calibration models, performance prediction models, cost prediction models, and sales volume prediction models can be fitted / trained to obtain the sensor. To improve the computational performance of the models, as a preferred implementation method, the input variables of each model are normalized before model calculation. For the performance prediction model, given performance indicators, the process parameters and environmental parameters can be constrained in reverse. Through the constrained process parameters and corresponding product parameters, possible costs and sales volumes can be further inferred. By evaluating the distance between the costs and sales volumes and the optimization objectives, the process parameters and product parameters can be updated in reverse. Through this iterative process, multi-objective parameter collaborative optimization of the sensor under constraints can be automatically and efficiently achieved.

[0067] Based on the above relationships, in one optional implementation, see Appendix Figure 3 Methods for optimizing sensor process parameters and product parameters include: S24. Initialize candidate solutions for process parameters and product parameters.

[0068] The candidate solution contains multiple sets of process parameters and product parameters. These multiple sets of process parameters and product parameters can be obtained through random initialization.

[0069] S25. Based on the candidate solutions, evaluate the performance indicators of the sensor under different environmental parameters, and select feasible solutions that meet the constraints from the candidate solutions.

[0070] As stated in the general requirements of step S2, the optimization of process parameters and product parameters requires predetermined performance indicators as constraints, and the performance prediction model constructed above must be used. (or other implementation methods) can deduce the performance indicators of each group of candidate solutions under different environmental parameters (corresponding to predetermined performance indicator items). Combinations that do not meet the predetermined performance indicators (i.e., constraints) do not need to be considered. From each group of candidate solutions, the process parameters and product parameters that meet the predetermined performance indicators are selected as feasible solutions for further optimization.

[0071] S26. Estimate the cost and / or sales volume of the sensor based on the feasible solution.

[0072] For example, the cost prediction models described above are used respectively. Or sales forecasting model Predict the cost for each of the selected feasible solutions. or sales .

[0073] S27. Based on the estimated cost and / or sales volume, update the candidate solutions with the objective of minimizing cost and / or maximizing sales volume.

[0074] The optimization objectives for the multi-objective parameters of the sensor are to minimize cost and / or maximize sales. Based on the estimated direction and magnitude of changes in cost and / or sales, candidate solutions (e.g., selected feasible solutions) are updated using the actual optimization algorithm employed (e.g., selection, crossover, and mutation in genetic algorithms, or the acquisition function of Bayesian optimization algorithms). Furthermore, when updating candidate solutions, conflicts between multi-objective parameters (i.e., cost and sales) are weighed, such as whether "cost reduction may lead to a decrease in performance margin" or "increased sales may require increased cost." Options that cause conflicting objective parameters are discarded during candidate solution updates.

[0075] According to the above design process, candidate solutions are updated iteratively until the stopping condition is met (such as reaching the maximum number of iterations, or the convergence of multiple objective parameters). At this point, the optimal objective parameters (or objective parameter set) that satisfy the constraints are obtained.

[0076] Through the above process, a closed loop of "model evaluation-optimization feedback" is achieved, which can realize the collaborative optimization of multiple objective parameters (optional) without the need for trial and error of physical mechanism models. This not only greatly improves the efficiency of parameter optimization, but also ensures that the optimal parameters can be found by chance, and avoids the occurrence of invalid solution sets.

[0077] For example, suppose the optimization objective is to minimize cost, which is inferred through a cost prediction model. The constraints are predetermined performance metrics, which are inferred through a performance prediction model. Then the optimization model for this example is expressed as: ; In the formula, This indicates the configuration of the trained model parameters. Cost prediction model; This indicates the configuration of the trained model parameters. Performance prediction model; This indicates that the environmental parameters are The expectations below.

[0078] By combining heuristic algorithms or Bayesian optimization algorithms with random sampling, the process parameters that best meet the performance indicators or have the lowest cost under the constraints of the performance indicators are found. and product parameters Combination. If the optimization objective is to maximize sales, or simultaneously minimize costs and maximize sales, then change the objective function to the corresponding optimization objective, keep the constraints unchanged, and use the same optimization algorithm to search for the optimal process parameters. and product parameters combination.

[0079] It should be noted that in the multi-objective parameter optimization method of this application, the various relationships (prediction models) are not independent but mutually influential. For example, adjusting process parameters to reduce costs may lead to a decrease in performance indicators, deviating from the scenario requirements (i.e., constraints); or adjusting product parameters to increase sales may lead to an increase in costs. Therefore, the various stages of the multi-objective parameter optimization method of this application influence each other and are optimized synchronously, which is something that no single-objective optimization method can achieve, nor is it simply a combination of two single optimization objectives.

[0080] In addition, as a preferred embodiment, this application also considers continuous optimization of the method: First, a complete method for optimizing multi-target parameters of nuclear radiation sensors is deployed to the target scenario.

[0081] Secondly, following the methodology, the first round of multi-target parameter optimization for the sensor was completed, resulting in a preliminary optimization scheme.

[0082] Then, the optimized sensor is actually manufactured and tested, and new performance data, cost data and market feedback data are collected.

[0083] Next, the newly collected data is added to the training dataset, and the various machine learning models are retrained to update the performance prediction model, cost prediction model, and sales prediction model.

[0084] Finally, based on the updated model, a new round of multi-objective parameter optimization is carried out to obtain an improved sensor design.

[0085] The aforementioned continuous optimization mechanism can be implemented periodically (e.g., quarterly). By establishing regular model updates and optimization cycles, it ensures that sensor designs are always based on the latest data and technological advancements, achieving continuous improvement.

[0086] To facilitate understanding of the present application, the embodiments of the present application also provide the following embodiments to illustrate the multi-target parameter optimization method for nuclear radiation sensors of the present application.

[0087] Suppose that the historical dataset collected from a certain type of nuclear radiation sensor contains the following information: Process parameters This includes sensor element fabrication process parameters (such as scintillator material type, photomultiplier tube size, filter circuit design, CZT wafer cutting accuracy (accuracy range ±0.01mm to ±0.05mm), ion implantation dose (10... 12 Up to 10 1 ions / cm 2 ), passivation layer thickness (50nm-200nm), circuit design and fabrication process parameters (such as circuit wiring spacing (0.1mm-0.5mm), component soldering accuracy (±0.05mm), shielding layer thickness (1-5mm)), and mass production process parameters (such as assembly tolerance range (±0.02mm to ±0.05mm)).

[0088] Product Parameters This includes the casing material, interface type, and display screen size.

[0089] Cost data, including raw material purchase prices and processing fees.

[0090] Environmental parameters Including temperature air humidity Background radiation intensity )wait.

[0091] Performance indicators Including energy resolution E / E Sensitivity S Detection efficiency η wait.

[0092] They also collected data on the parameters, prices, and sales volume of similar products on the market.

[0093] First, three machine learning models were trained based on this historical dataset: (1) Performance prediction model The random forest algorithm is used, with process parameters as the input variables. With environmental parameters The output variable is E / E , S , η The performance prediction model after training on the test set It reached 0.74. (2) Cost prediction model The XGBoost algorithm is used, with process parameters as the input variable. With product parameters The output is the cost. c The root mean square error of the cost prediction model after training is 85 yuan. (3) Sales prediction model A neural network is used, with product parameters as the input variable. and price p The output variable is sales volume. s The sales prediction model after training performs well on the test set. It is 0.69.

[0094] Next, set the optimization objective: to meet the performance index constraints ( E / E ≤5%, S ≥0.5cps / Bq, η Minimize cost under the premise of ≥30% c (The other two optimization objectives can be modified similarly). A genetic algorithm is used for optimization, with a population size of 200 and 100 generations.

[0095] Optimization process: Randomly generate the initial population, and utilize... Assess each sample in the worst environmental combination (T=50) M=90%, R=10 µSv / h Under the given performance, individuals that meet the constraints are selected; using Calculate the cost; based on the cost, perform selection, crossover, and mutation to generate a new generation of population. After 100 generations, obtain the optimal parameter combination.

[0096] Optimization result: The cost corresponding to the optimal parameter combination is Yuan, compared to the original design cost c 0 = 200 yuan, a reduction of 27.5%; Optimized process parameters include: CZT wafer cutting accuracy ±0.02mm (reduced accuracy requirements), ion implantation dose 8×10 12 ions / cm 2 (Optimized dosage), circuit wiring spacing 0.3mm (relaxed spacing requirements), assembly tolerance range ±0.03mm (relaxed tolerances). Within the environmental range... , , Within this range, the predicted performance index value is E / E =4.7%, S=0.53 cps / Bq , η =31.5%, fully meeting the assessment indicators. This optimization result was achieved through multi-model collaborative optimization, involving the coordinated adjustment of multiple process parameters, while ensuring performance robustness across the entire environmental range, which is difficult to obtain through conventional experience or adjustment of a single model.

[0097] Finally, the optimized parameter combination was fabricated into a physical object and tested. The actual cost was 148 yuan, and the performance indicators were basically consistent with the predictions, thus verifying the effectiveness of the method in this application.

[0098] Optimization Process Analysis: The optimization process in this embodiment demonstrates that the optimization method does not adjust individual parameters in isolation, but rather finds a synergistic relaxation combination among "CZT wafer dicing accuracy," "ion implantation dose," "circuit wiring spacing," and "assembly tolerance." For example, by moderately relaxing the dicing accuracy requirement (from ±0.01mm to ±0.02mm) while optimizing the ion implantation dose and relaxing the wiring spacing, significant cost savings can be achieved while ensuring overall performance. This multi-objective parameter linkage optimization reflects the value of multi-model synergy (performance prediction model and cost prediction model) in the method of this application for complex engineering problems, an objective that is difficult to achieve through conventional experience or adjustment of a single model.

[0099] Comparative Explanation (Simulation): To highlight the advantages of the method of this application, comparative experimental simulations were also conducted in the embodiments of this application: Comparison with Option 1: Adjusting target parameters solely based on engineer experience (keeping all parameters within the "high precision" range).

[0100] Comparison Option 2: Focusing solely on minimizing costs while ignoring performance constraints.

[0101] The results showed that the cost of Scheme 1 was 198 yuan, a reduction of only 1%, and it did not fully utilize the synergistic effect between parameters; although the cost of Scheme 2 was reduced to 130 yuan, it was less effective in harsh environments. E / E The current performance index is 6.2%, which does not meet the performance requirements. However, the method described in this application achieves a significant cost reduction (27.5%) while ensuring that the overall environmental performance index meets the requirements, demonstrating the comprehensive advantages of multi-objective synergistic optimization.

[0102] Based on the concept of this application, an embodiment of this application also provides a multi-target parameter optimization device for a nuclear radiation sensor. This device includes a processor and a storage medium. The storage medium stores computer instructions, and the processor executes these computer instructions to perform the multi-target parameter optimization method for a nuclear radiation sensor according to any of the above embodiments.

[0103] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the nuclear radiation sensor multi-target parameter optimization method according to any of the above embodiments.

[0104] In addition, this application also provides a computer program product, which includes a computer program that, when run by a processor, executes the nuclear radiation sensor multi-target parameter optimization method according to any of the above embodiments.

[0105] Furthermore, embodiments of this application also provide a multi-target parameter optimization system for nuclear radiation sensors, the system comprising: The user interaction module receives external data to pass to the processing module, or outputs data from the processing module to the outside. External data may include, for example, historical data collected or predefined performance indicators; output data may include, for example, optimized multi-objective parameters, i.e., a combination of process parameters and product parameters.

[0106] The processing module is used to execute the multi-target parameter optimization method for nuclear radiation sensors according to any of the above embodiments.

[0107] For example, the aforementioned nuclear radiation sensor multi-target parameter optimization system is deployed on the web and adopts a front-end and back-end separation architecture. The front-end is implemented using HTML / CSS / JavaScript, and the back-end uses the Python Flask framework to provide API services.

[0108] See appendix Figure 4 The processing module includes the following units: 1. Data Upload Unit: Users can upload sensor test data (CSV or Excel format) by dragging and dropping or selecting files. The system automatically performs preprocessing operations such as data cleaning, outlier handling, and normalization.

[0109] 2. Statistical Analysis Unit: Performs descriptive statistical analysis (mean, variance, distribution, etc.) on the uploaded data and visualizes it through interactive charts (Chart.js).

[0110] 3. Regression Model Unit: Users can choose from four prediction models (parameter calibration model, performance prediction model, cost prediction model, and sales prediction model) and select a model algorithm (linear regression, neural network, random forest, XGBoost, etc.). The system supports model training and evaluation (R... 2 (Indicators such as RMSE) and their preservation.

[0111] 4. Parameter Optimization Unit: Users can set optimization objectives (minimize cost and / or maximize sales, etc.) and constraints (energy resolution, sensitivity, detection efficiency, etc.), and select optimization algorithms (evolutionary algorithm, particle swarm optimization, Bayesian optimization, etc.). The system supports multi-dimensional model evaluation (physical model or training model) and returns the optimized parameter combination.

[0112] 5. Results visualization unit (can be a user interaction module): Provides charts to display statistical analysis results, model prediction effects, convergence curves of the optimization process, etc.

[0113] The system operation process is as follows: (1) The user uploads sensor test data; (2) The system automatically preprocesses the data and displays the statistical analysis results; (3) The user selects a prediction model and trains it, and the system evaluates the model performance; (4) The user sets the optimization objective and constraints, and the system runs the optimization algorithm; (5) The system returns the optimized parameter combination and provides a visual report.

[0114] Figure 4The diagram shows the interface of the aforementioned web-based multi-objective parameter optimization system for nuclear radiation sensors. The system's functional module layout and operational logic are centered on user interaction. Its design revolves around a complete workflow of "data upload - preprocessing - statistical analysis - model building - parameter optimization," intuitively showcasing the system's ease of use and automation. The core functional modules of the interface include a data upload unit, a statistical analysis unit, a regression model unit, a parameter optimization unit (not shown), and a result visualization unit (not shown). The data upload unit allows users to upload sensor test data via drag-and-drop or click-to-browse, supporting CSV and Excel formats. File size is typically limited, for example, to no more than 10MB. Uploaded data must include core data on environmental parameters, process parameters, and sensor performance indicators. The data upload unit has built-in data preprocessing options, allowing users to perform operations such as data cleaning, outlier removal, and data normalization on the uploaded data. It can also perform feature extraction and statistical analysis, providing high-quality data support for subsequent modeling. The regression model unit... The model unit supports users in building various types of regression models based on preprocessed data, covering four categories: performance prediction based on environmental parameters, performance prediction based on process and environmental parameters, cost prediction based on process and product parameters, and sales prediction based on product parameters and selling price. Users can choose algorithms such as linear regression, neural networks, random forests, and XGBoost for model training and optimization. The parameter optimization unit allows users to set performance constraints and optimization objectives such as cost minimization and / or sales maximization, and select stochastic optimization algorithms such as evolutionary algorithms and particle swarm optimization to perform parameter optimization. Results visualization presents core information such as statistical analysis results, model prediction accuracy, convergence curves of the optimization process, and optimal parameter combinations through interactive charts.

[0115] This system automates and visualizes the entire process of the proposed solution, making it suitable for engineers, researchers, and decision-makers to quickly verify sensor performance and optimize parameters, thereby lowering the R&D threshold and improving decision-making efficiency.

[0116] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A method for optimizing multi-objective parameters of a nuclear radiation sensor, characterized in that, include: Perform parameter calibration on the sensor to adapt to various scenarios; Using the predetermined performance indicators of the calibrated sensor as constraints, and with the goal of minimizing cost and / or maximizing sales volume, the process parameters and product parameters of the sensor are optimized; wherein, the performance indicators are determined by the process parameters and environmental parameters of the sensor; the cost is determined by the process parameters and product parameters of the sensor; and the sales volume is determined by the product parameters and price parameters of the sensor.

2. The multi-objective parameter optimization method for nuclear radiation sensors as described in claim 1, characterized in that, Perform parameter calibration on the sensor to adapt to various scenarios, including: Under the action of a standard radiation source, the performance index test data of the sensor are collected under various scenarios and different environmental parameters. The test data is preprocessed; Establish the correlation between the sensor's performance indicators and usage scenarios and environmental parameters to infer the sensor's performance indicators under different scenarios and combinations of environmental parameters.

3. The multi-objective parameter optimization method for nuclear radiation sensors as described in claim 1, characterized in that, The reasoning method for the performance indicators includes: The performance prediction model is trained and inferred based on the current process parameters and environmental parameters of the sensor to obtain the sensor's performance index. The performance prediction model is obtained by training a machine learning model based on different process parameters, environmental parameters and corresponding performance indexes of the sensor collected in history.

4. The multi-objective parameter optimization method for nuclear radiation sensors as described in claim 1, characterized in that, The reasoning method for the cost includes: The cost of the sensor is obtained by reasoning based on the current process parameters and product parameters of the sensor using a trained cost prediction model. The cost prediction model is obtained by training a machine learning model based on different process parameters, product parameters and corresponding costs of the sensor collected in history.

5. The method for optimizing multi-objective parameters of a nuclear radiation sensor as described in claim 1, characterized in that, The reasoning method for the sales figures includes: The sales volume of the sensor is obtained by inferring based on the product parameters and price parameters of the sensor currently being used by the trained sales volume prediction model. The sales volume prediction model is obtained by training a machine learning model based on different product parameters, price parameters and corresponding sales volumes of the sensor collected in history.

6. The method for optimizing multi-objective parameters of a nuclear radiation sensor as described in any one of claims 1-5, characterized in that, Methods for optimizing the process parameters and product parameters of the sensor include: Candidate solutions for initializing process parameters and product parameters; Based on the candidate solutions, the performance indicators of the sensor under different environmental parameters are evaluated, and feasible solutions that meet the constraints are selected from the candidate solutions. Estimate the cost and / or sales volume of the sensor based on the feasible solution; The candidate solutions are updated based on estimated costs and / or sales, with the objective of minimizing costs and / or maximizing sales. This process continues until the stopping condition is met.

7. A multi-target parameter optimization device for a nuclear radiation sensor, characterized in that, It includes a processor and a storage medium, the storage medium storing computer instructions, the processor executing the computer instructions to perform the multi-objective parameter optimization method for nuclear radiation sensors as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, performs the multi-objective parameter optimization method for nuclear radiation sensors as described in any one of claims 1-6.

9. A computer program product, characterized in that, It includes a computer program, which, when executed by a processor, performs the multi-objective parameter optimization method for nuclear radiation sensors as described in any one of claims 1-6.

10. A multi-target parameter optimization system for a nuclear radiation sensor, characterized in that, include: The user interaction module is used to receive external data and pass it to the processing module, or to output data output by the processing module to the outside. The processing module is used to execute the multi-objective parameter optimization method for nuclear radiation sensors as described in any one of claims 1-6.

Images