Multi-objective optimization system for high-temperature alloy deep hole drilling process parameters

Abstract

The invention relates to the technical field of high-temperature alloy machining, and discloses a high-temperature alloy deep hole drilling machining process parameter multi-objective optimization system which comprises a multi-source heterogeneous data synchronous acquisition module, a physical information neural network digital twin proxy modeling module and a dynamic Pareto front edge low-delay multi-objective optimization module which are in communication connection in sequence. Through a multi-source heterogeneous data synchronous acquisition module, an integrated piezoelectric dynamometer, a high-frequency current transformer and other special sensors, key signals such as cutting force, main shaft power, cooling liquid pressure, drilling depth and the like are synchronously acquired at a high sampling rate not lower than 10kHz, timestamp alignment of multi-source data is completed at the same time, and standardized JSON structured data is output; according to the design, the problems of asynchronous data acquisition, disordered format and insufficient precision in the prior art are solved, high-precision and high-reliability data source support is provided for subsequent PINN modeling and multi-target optimization, and the operation basis of the whole system is ensured.

Description

Technical Field

[0001] This invention relates to the field of high-temperature alloy processing technology, specifically a multi-objective optimization system for process parameters of deep hole drilling of high-temperature alloys. Background Technology

[0002] High-temperature alloys (such as Inconel 718 and GH4169) are widely used in the manufacture of core components for aero-engines, such as turbine disks and combustion chambers, due to their excellent high-temperature resistance, corrosion resistance, and high strength. Deep hole drilling (where the hole diameter D and hole depth L satisfy L>5D) is a critical process in the machining of these components, with typical applications including the machining of bolt holes in aero-engine turbine disks and cooling holes in blades. Deep hole drilling takes place in a semi-enclosed space with a complex machining environment. The cutting zone exhibits a strong coupling effect of thermodynamics, fluid mechanics, and tool wear, making the rational determination of process parameters a key challenge in ensuring machining quality, efficiency, and tool life.

[0003] Currently, the determination of process parameters for deep hole drilling of high-temperature alloys mainly relies on traditional experience manuals, offline cutting tests, or simple trial cutting and debugging methods. This method has many insurmountable defects and cannot meet the high precision, high efficiency, and high stability requirements of modern intelligent manufacturing, as follows: Insufficient accuracy and synchronization of multi-source data acquisition: In the current machining process, the acquisition of key state parameters such as cutting force, spindle power, and coolant pressure mostly adopts a single sensor and low sampling rate (below 10kHz), and the timestamp alignment of multi-source data is not achieved. It is impossible to accurately capture the dynamic state of deep hole drilling as the drilling depth changes, resulting in a lack of reliable data support for subsequent modeling and optimization. At the same time, the acquired data is mostly in unstructured format, which is difficult to use directly for subsequent algorithm analysis and model training. Cutting condition prediction models suffer from poor robustness and insufficient generalization ability: Existing technologies for predicting tool wear, cutting temperature, and surface roughness are mostly purely data-driven black-box models (such as BP neural networks and response surface methodology) or single-mechanism models, failing to deeply integrate physical laws with data-driven models. Purely data-driven models lack physical constraints, leading to exponential divergence in prediction errors when faced with unseen tool wear conditions or variations in drilling depth. Single-mechanism models, on the other hand, cannot accurately adapt to the strong coupling characteristics of multi-physics fields in deep hole drilling, making high-precision prediction difficult. Multi-objective optimization suffers from poor real-time performance and insufficient adaptability: Deep hole drilling of high-temperature alloys requires simultaneous three-dimensional multi-objective optimization to maximize material removal rate (MRR), minimize surface roughness (Ra), and minimize tool wear (VB). This optimization problem is a typical non-convex Pareto optimization problem. Existing technologies mostly employ traditional genetic algorithms (such as NSGA-II) or offline optimization methods, failing to optimize for the machining cycle time requirement of <30s in deep hole drilling. A single optimization can take up to minutes, making real-time dynamic optimization impossible. Furthermore, the lack of Chebyshev scalarization mechanisms and reference point guidance strategies makes it difficult to balance optimization accuracy and efficiency, and it cannot dynamically adapt to the machining state fluctuations caused by changes in drilling depth. Lack of closed-loop control capability and poor system architecture adaptability: Existing deep hole drilling is mostly open-loop or semi-closed-loop control, lacking a complete closed-loop link from data acquisition, state prediction, parameter optimization to parameter distribution; even if some closed-loop control schemes exist, they do not use standardized communication protocols (such as TCP / IP) to achieve efficient interaction with the CNC system, and cannot achieve real-time interpolation coverage of process parameters. In addition, existing systems are mostly monolithic architectures, do not adopt microservice distributed design, cannot achieve high concurrency and high reliability operation of multiple modules, and are difficult to adapt to the large-scale deployment requirements in industrial scenarios. Summary of the Invention

[0004] This invention provides a multi-objective optimization system for the process parameters of deep hole drilling of high-temperature alloys to solve the problems mentioned in the background art.

[0005] This invention provides the following technical solution: a multi-objective optimization system for deep hole drilling process parameters of high-temperature alloys, comprising a multi-source heterogeneous data synchronous acquisition module, a physical information neural network digital twin agent modeling module, a dynamic Pareto front low-latency multi-objective optimization module, and an adaptive decision-making and closed-loop control module connected in sequence. The multi-source heterogeneous data synchronous acquisition module is configured to synchronously acquire cutting force signals, spindle power signals, coolant pressure signals and drilling depth signals at a sampling rate of not less than 10kHz, and complete the timestamp alignment of multi-source data to output structured machining status data. Through the collaborative design of high-frequency sensors and edge computing units, synchronous acquisition and standardized processing of multi-source signals are achieved. The high sampling rate of 10kHz can accurately capture instantaneous cutting force fluctuations, spindle power mutations, and coolant pressure dynamic changes during deep hole drilling, avoiding the loss of critical status information due to low sampling frequency. The multi-source data timestamp alignment adopts a dual mechanism of hardware clock synchronization and software timestamp calibration to eliminate clock deviations between different sensors and the CNC system, ensuring that the output structured machining status data has high temporal consistency. This provides accurate and reliable data source support for subsequent model training and optimization, solving the technical pain points of insufficient accuracy and poor synchronization in traditional data acquisition.

[0006] The physical information neural network digital twin agent modeling module is configured to use spindle speed, feed rate, high-pressure internal cooling pressure and current drilling depth as input vectors, and tool wear, cutting temperature and surface roughness as output vectors. The network training is completed by using a composite loss function that integrates data loss terms and physical residual terms. The physical residual terms are embedded with the Oxley cutting mechanics analytical equation and the one-dimensional transient heat conduction Fourier partial differential equation. The dynamic Pareto front low-latency multi-objective optimization module is configured to construct a three-dimensional objective optimization space by maximizing material removal rate, minimizing surface roughness, and minimizing tool wear. Based on the output of the digital twin surrogate model as the fitness evaluation function, the improved non-dominated sorting genetic algorithm NSGA-Ⅲ is used to dynamically solve the Pareto optimal solution set. The adaptive decision-making and closed-loop control module is configured to select the optimal combination of process parameters from the Pareto optimal solution set and send the optimal process parameters to the CNC system through the communication interface to realize real-time closed-loop adjustment of the deep hole drilling process.

[0007] As a preferred technical solution of the present invention, the composite loss function is expressed as: Loss = α・Loss_data + β・Loss_physics, where Loss_data is the mean square error between the network prediction value and the measured value, Loss_physics is the physical residual term, and α and β are preset weighting coefficients.

[0008] As a preferred embodiment of the present invention, the physical information neural network digital twin agent modeling module is further configured to: calculate the partial derivative of the cutting temperature output by the network with respect to the drilling depth, and apply an exponential penalty factor to the prediction results that violate the high-pressure fluid convection heat transfer boundary conditions, and include the penalty results in the physical residual term.

[0009] As a preferred technical solution of the present invention, the multi-source heterogeneous data synchronous acquisition module includes a piezoelectric force gauge, a high-frequency current transformer, a coolant pressure sensor, and a CNC pose analysis unit. The output structured machining status data is a standardized JSON object, which includes timestamp, drilling depth, three-dimensional cutting force, spindle power, and coolant pressure fields.

[0010] As a preferred technical solution of the present invention, the dynamic Pareto front low-latency multi-objective optimization module integrates Chebyshev scalarization mechanism and reference point guidance strategy, and the calculation delay of a single Pareto front solution does not exceed 500ms.

[0011] As a preferred technical solution of the present invention, the dynamic Pareto front low-latency multi-objective optimization module is configured to access the current drilling depth and structured processing status data in real time during the evolutionary iteration process, dynamically update the fitness evaluation, and generate the optimal parameter solution set for the next processing cycle.

[0012] As a preferred embodiment of the present invention, the adaptive decision-making and closed-loop control module is configured to calculate the Euclidean distance between each candidate solution in the Pareto optimal solution set and the ideal point in the normalized target space, and select the solution with the smallest distance as the optimal combination of process parameters for final execution.

[0013] As a preferred embodiment of the present invention, the adaptive decision-making and closed-loop control module adopts the Ethernet TCP / IP protocol to write the optimal process parameters into the macro variable register of the CNC system, thereby realizing real-time interpolation coverage of spindle speed, feed rate and cooling pressure.

[0014] As a preferred technical solution of the present invention, the system adopts a microservice distributed architecture based on Spring Boot, wherein: the multi-source heterogeneous data synchronization and acquisition module is deployed in the form of a microservice on the EdgeXFoundry edge computing framework; the physical information neural network digital twin agent modeling module is deployed on a TensorFlowServing container cluster, providing a high-concurrency inference interface.

[0015] As a preferred embodiment of the present invention, the dynamic Pareto front low-latency multi-objective optimization module is configured with a Redis in-memory database middleware for high-speed caching of processing state data and parallel computation acceleration of multi-objective evolutionary optimization.

[0016] The present invention has the following beneficial effects: 1. This high-temperature alloy deep hole drilling process parameter multi-objective optimization system integrates piezoelectric force gauges, high-frequency current transformers and other special sensors through a multi-source heterogeneous data synchronous acquisition module. It synchronously acquires key signals such as cutting force, spindle power, coolant pressure and drilling depth at a high sampling rate of not less than 10kHz. At the same time, it completes the timestamp alignment of multi-source data and outputs standardized JSON structured data. This design solves the problems of asynchronous data acquisition, chaotic format, and insufficient accuracy in existing technologies, providing high-precision and high-reliability data source support for subsequent PINN modeling and multi-objective optimization, and ensuring the basic operation of the entire system.

[0017] 2. The multi-objective optimization system for deep hole drilling process parameters of high-temperature alloys constructs a digital twin surrogate model by using a physical information neural network (PINN). It integrates the Oxley cutting mechanics analytical equation with the one-dimensional transient heat conduction Fourier partial differential equation into a physical residual term, designs a composite loss function that integrates data loss and physical residual, and applies an exponential penalty factor to the prediction results that violate the convection heat transfer boundary conditions. This design breaks through the black-box limitations of traditional pure data-driven models, solidifies physical laws into the model's underlying layer, and enables the model to have stronger generalization ability. When faced with unseen tool wear conditions or changes in drilling depth, the root mean square error (RMSE) of the prediction is lower than that of traditional BP neural networks, and it can accurately predict tool wear, cutting temperature and surface roughness in real time.

[0018] 3. This multi-objective optimization system for deep hole drilling process parameters of high-temperature alloys adopts an improved NSGA-Ⅲ algorithm through a dynamic Pareto front low-delay multi-objective optimization module, integrates Chebyshev scalarization mechanism and reference point guidance strategy, and uses the output of the PINN model as the fitness evaluation function to achieve dynamic optimization of the three-dimensional target space (Max(MRR), Min(Ra), Min(VB)). This design reduces the latency of a single Pareto front solution to less than 500ms, solving the problems of high computational overhead and long optimization time in existing algorithms. At the same time, it can access the current drilling depth and processing status data in real time during the evolution and iteration process, dynamically update the optimal parameter solution set, perfectly adapt to the processing cycle requirements of deep hole drilling, and realize dynamic evolution optimization.

[0019] 4. This multi-objective optimization system for deep hole drilling of high-temperature alloys uses an adaptive decision-making and closed-loop control module to select the optimal process parameters from the Pareto optimal solution set using the Euclidean distance decision method. The parameters are then written into the macro variable register of the CNC system via the Ethernet TCP / IP protocol to achieve real-time interpolation coverage of spindle speed, feed rate, and cooling pressure. This design constructs a complete closed-loop chain of "data acquisition - state prediction - parameter optimization - parameter distribution - real-time adjustment", which solves the problems of lack of closed-loop control and lag in parameter adjustment in existing technologies, ensuring that the processing can adapt to state changes in real time and improve processing stability.

[0020] 5. This multi-objective optimization system for deep hole drilling process parameters of high-temperature alloys adopts a SpringBoot microservice distributed architecture, deploys the multi-source data acquisition module on the EdgeX Foundry edge computing framework, deploys the PINN proxy model on the TensorFlow Serving container cluster, and configures the optimization module with Redis in-memory database middleware. This design enables independent deployment of multiple modules, high-concurrency operation, and flexible expansion. The introduction of Redis middleware enables high-speed caching of processing status data and parallel computing acceleration of the optimization process, further reducing system response latency, improving system reliability and scalability, and adapting to the needs of large-scale industrial deployment. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the overall workflow of the present invention. Figure 2 This is a flowchart of the physical information neural network modeling process of the present invention; Figure 3 This is a flowchart of the dynamic Pareto multi-objective optimization decision-making process of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1-3 The high-temperature alloy deep hole drilling process parameter multi-objective optimization system includes a multi-source heterogeneous data synchronous acquisition module, a physical information neural network digital twin agent modeling module, a dynamic Pareto front low-latency multi-objective optimization module, and an adaptive decision and closed-loop control module, which are connected in sequence. The above four modules form a complete closed-loop architecture of "perception-modeling-optimization-control". The modules interact with each other through real-time communication links, abandoning the traditional offline optimization mode and realizing online collaboration of the entire process from processing status perception to adaptive adjustment of process parameters. It is suitable for high-temperature alloy deep hole drilling conditions where the hole diameter D and hole depth L meet L>5D, and can effectively solve the process optimization problem caused by strong coupling of multiple physical fields in a semi-enclosed cutting space.

[0024] The multi-source heterogeneous data synchronous acquisition module is configured to synchronously acquire cutting force signals, spindle power signals, coolant pressure signals and drilling depth signals at a sampling rate of not less than 10kHz, and complete the timestamp alignment of multi-source data to output structured machining status data. Using a high sampling rate of no less than 10kHz, it can fully capture transient features such as sudden changes in cutting force, spindle load fluctuations, and cooling pressure pulsations during deep hole drilling, avoiding the loss of key working condition information caused by low-frequency sampling. The time stamp alignment mechanism eliminates the timing deviation between different sensors and CNC systems, ensuring that multi-source data are fused under the same time reference, providing time-consistent and highly reliable structured machining status data for subsequent modeling and optimization.

[0025] The physical information neural network digital twin agent modeling module is configured to use spindle speed, feed rate, high-pressure internal cooling pressure and current drilling depth as input vectors, and tool wear, cutting temperature and surface roughness as output vectors. The network training is completed by using a composite loss function that integrates data loss terms and physical residual terms. The physical residual terms are embedded with the Oxley cutting mechanics analytical equation and the one-dimensional transient heat conduction Fourier partial differential equation. In the above structure, process parameters and machining position are used as model inputs, and machining quality, tool condition and thermal condition are used as outputs to construct a digital twin mapping relationship for the deep hole drilling process. By embedding cutting mechanics and heat conduction physical equations into the loss function, the network prediction results no longer rely solely on data fitting, but simultaneously satisfy the laws of cutting physics, significantly improving the prediction stability and generalization ability of the model under large depth-to-diameter ratio conditions, and overcoming the problem of prediction distortion in traditional data-driven models.

[0026] The dynamic Pareto front low-latency multi-objective optimization module is configured to construct a three-dimensional objective optimization space by maximizing material removal rate, minimizing surface roughness, and minimizing tool wear. The output of the digital twin surrogate model is used as the fitness evaluation function, and the improved non-dominated sorting genetic algorithm NSGA-Ⅲ is used to dynamically solve the Pareto optimal solution set. In the above structure, the material removal rate, surface roughness, and tool wear constitute mutually constraining three-dimensional optimization objectives, which are in line with the actual production needs of deep hole drilling of high-temperature alloys. The digital twin proxy model is used as a rapid fitness evaluation to replace finite element simulation and physical trial cutting, which greatly reduces the computation time. By improving the NSGA-Ⅲ algorithm, the optimal frontier is quickly approximated in the non-convex target space, and a set of Pareto optimal solutions that take into account efficiency, quality and tool life is obtained.

[0027] The adaptive decision-making and closed-loop control module is configured to select the optimal combination of process parameters from the Pareto optimal solution set and send the optimal process parameters to the CNC system through the communication interface to realize real-time closed-loop adjustment of the deep hole drilling process. This module is responsible for optimization result decision-making and instruction execution. By making optimal decisions on the Pareto solution set, it obtains a set of comprehensive optimal process parameters. Through the communication interface with the CNC system, it realizes real-time parameter distribution and dynamic adjustment, enabling the machining process to correct process parameters in real time according to changes in drilling depth and cutting state, forming an online closed-loop optimization, and avoiding the problem of mismatch between traditional offline parameter schemes and actual working conditions.

[0028] In a preferred embodiment, the composite loss function is expressed as: Loss = α・Loss_data + β・Loss_physics, where Loss_data is the mean square error between the network prediction and the measured value, Loss_physics is the physical residual term, and α and β are preset weighting coefficients. In the above structure, the weighting coefficients α and β are used to balance the data fitting accuracy and the physical constraint strength, respectively. α is used to control the degree of model fitting to the measured data, and β is used to control the degree of observability of the prediction results to the physical laws of cutting. By reasonably allocating the weights, the model can have both high data accuracy and strong physical rationality, and is suitable for deep hole drilling scenarios with different grades of high temperature alloys and different types of cutting tools.

[0029] In a preferred embodiment, the physical information neural network digital twin agent modeling module is further configured to: calculate the partial derivative of the cutting temperature output by the network with respect to the drilling depth, and apply an exponential penalty factor to the prediction results that violate the high-pressure fluid convection heat transfer boundary conditions, and include the penalty results in the physical residual term; In the above structure, by constraining the gradient change of cutting temperature along the drilling depth, the temperature prediction conforms to the convective heat transfer characteristics under high-pressure internal cooling conditions; the exponential penalty factor can impose stronger constraints on the prediction results that deviate significantly from the physical boundary conditions, forcing the network to gradually converge to the output that conforms to the real heat transfer law during the training process, thereby further improving the prediction accuracy of key indicators such as cutting temperature and tool wear.

[0030] In a preferred embodiment, the multi-source heterogeneous data synchronous acquisition module includes a piezoelectric force gauge, a high-frequency current transformer, a coolant pressure sensor, and a CNC pose analysis unit. The output structured machining status data is a standardized JSON object, which includes timestamp, drilling depth, three-dimensional cutting force, spindle power, and coolant pressure fields. In the above structure, the piezoelectric force gauge is used to collect high-precision three-dimensional cutting force, the high-frequency current transformer reflects the changes in spindle load in real time, the coolant pressure sensor monitors the working status of the internal cooling system, and the CNC pose analysis unit obtains the real-time drilling depth. The information collected by each hardware is uniformly encapsulated in JSON format, which facilitates cross-module parsing and calling between edge nodes, cloud models and optimization algorithms, and improves system compatibility and scalability.

[0031] In a preferred embodiment, the dynamic Pareto front low-latency multi-objective optimization module integrates Chebyshev scalarization mechanism and reference point guidance strategy, with a single Pareto front solution computation latency of no more than 500ms; In the above structure, Chebyshev scalarization is used to balance the conflict between multiple objectives, and the reference point guidance strategy improves the convergence speed and distribution uniformity of the solution set. The combination of the two enables the improved NSGA-III to significantly improve computational efficiency while ensuring optimization quality. The single solution delay is controlled within 500ms, which can meet the cycle time requirements of continuous deep hole drilling and realize true online dynamic optimization.

[0032] In a preferred embodiment, the dynamic Pareto front low-latency multi-objective optimization module is configured to access the current drilling depth and structured processing status data in real time during the evolutionary iteration process, dynamically update the fitness evaluation, and generate the optimal parameter solution set for the next processing cycle. In the above structure, the optimization process is linked with the drilling depth in real time. It can dynamically update the target evaluation and optimization direction according to the changes in working conditions such as the accumulation of cutting heat, changes in chip removal conditions, and increased tool wear caused by the increase in hole depth. It continuously generates process parameters that are suitable for the next drilling stage, avoiding quality fluctuations and abnormal tool wear caused by using fixed process parameters.

[0033] In a preferred embodiment, the adaptive decision-making and closed-loop control module is configured to calculate the Euclidean distance between each candidate solution in the Pareto optimal solution set and the ideal point in the normalized target space, and select the solution with the smallest distance as the optimal combination of process parameters to be executed. In the above structure, the differences in the dimensions of different objectives are eliminated by normalization, and the Euclidean distance is used to measure the closeness of the candidate solution to the ideal optimization objective. The solution with the smallest distance is selected as the execution parameter. The optimal combination of process parameters with comprehensive performance can be quickly determined among multiple non-dominated solutions. The decision-making logic is simple and efficient, and the real-time performance of the system is guaranteed.

[0034] In a preferred embodiment, the adaptive decision-making and closed-loop control module uses the Ethernet TCP / IP protocol to write the optimal process parameters into the macro variable register of the CNC system, thereby achieving real-time interpolation coverage of spindle speed, feed rate and cooling pressure. In the above structure, the TCP / IP protocol is used to ensure the stability and real-time performance of data transmission. The process parameters are seamlessly integrated by writing them into the CNC system macro variables. The speed, feed and cooling pressure can be dynamically adjusted without modifying the machining program, and the real-time interpolation correction of the machining process is realized, which improves the automation and intelligence level of the deep hole drilling process.

[0035] In a preferred embodiment, the system adopts a microservice distributed architecture based on Spring Boot, wherein: the multi-source heterogeneous data synchronization and acquisition module is deployed as a microservice on the EdgeXFoundry edge computing framework; the physical information neural network digital twin agent modeling module is deployed on the TensorFlowServing container cluster, providing a high-concurrency inference interface; In the above structure, the microservice architecture decouples the functional modules, making it easy to upgrade, expand and maintain independently; the EdgeXFoundry edge framework reduces data acquisition and transmission latency and improves on-site response speed; TensorFlowServing containerized deployment supports high-concurrency model inference and can provide proxy model services for multiple processing devices at the same time, improving the overall system throughput and engineering applicability.

[0036] In a preferred embodiment, the dynamic Pareto front low-latency multi-objective optimization module is configured with a Redis in-memory database middleware for high-speed caching of processing state data and parallel computation acceleration of multi-objective evolutionary optimization; In the above structure, the Redis in-memory database enables low-latency reading and writing of processing status data, avoiding computational delays caused by frequent disk I / O; at the same time, it supports multi-threaded parallel computing, which can execute individual fitness evaluation in evolutionary algorithms in parallel, further shortening the optimization time and ensuring that the system stably meets the low-latency optimization requirements during continuous processing.

[0037] Working principle: In the deep hole drilling process of high-temperature alloys, the core working logic is based on real-time sensing, physical constraint modeling, dynamic multi-objective optimization, and closed-loop parameter adjustment. The specific working principle is as follows: Multi-source real-time sensing of working conditions: The multi-source heterogeneous data synchronous acquisition module uses a piezoelectric force gauge, a high-frequency current transformer, a pressure sensor and a CNC pose analysis unit to synchronously acquire cutting force, spindle power, cooling pressure and drilling depth signals at a sampling rate of no less than 10kHz. After being timestamped, the data is converted into standardized JSON structured data to provide the system with real-time and unified machining status input. Cutting state prediction under physical information constraints: The physical information neural network digital twin agent modeling module takes spindle speed, feed rate, cooling pressure, and drilling depth as inputs, and outputs tool wear, cutting temperature, and surface roughness through a multilayer perceptron network. Network training employs a composite loss function, which weights and fuses the data fitting loss with physical residuals embedded in Oxley cutting mechanics and one-dimensional transient heat conduction equations. An exponential penalty is applied to cases where the cutting temperature gradient violates the convective heat transfer boundary conditions, ensuring that the model output simultaneously satisfies data accuracy and physical laws, achieving stable and reliable state prediction under large aspect ratios. Low-latency dynamic multi-objective optimization: The dynamic Pareto front low-latency multi-objective optimization module constructs a three-dimensional target space with the optimization objectives of maximizing material removal rate, minimizing surface roughness, and minimizing tool wear. The output of the digital twin agent model is used as a fast fitness evaluation function, and an improved NSGA-III algorithm integrating Chebyshev scalarization and reference point guidance is used for evolutionary optimization. During the optimization process, drilling depth and machining status data are accessed in real time, and the fitness evaluation is dynamically updated to generate the Pareto optimal solution set. The Redis in-memory database realizes high-speed data caching and parallel computing acceleration, so that the latency of a single optimization does not exceed 500ms. Adaptive decision-making and CNC closed-loop execution: The adaptive decision-making and closed-loop control module normalizes the Pareto optimal solution set, calculates the Euclidean distance between each candidate solution and the ideal point, and selects the parameter combination with the smallest distance as the optimal process parameters; the parameters are written into the macro variable register of the CNC system through the Ethernet TCP / IP protocol, and the spindle speed, feed rate and cooling pressure are covered and adjusted in real time to realize the closed-loop adaptive adjustment of the drilling process; The system adopts a SpringBoot microservice distributed architecture, with each module deployed independently and running collaboratively, forming a complete intelligent optimization closed loop from perception, modeling, optimization to control, which significantly improves the processing efficiency, surface quality and tool life of deep hole drilling of high-temperature alloys.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. A multi-objective optimization system for deep hole drilling process parameters of high-temperature alloys, characterized in that: It includes a multi-source heterogeneous data synchronous acquisition module that is connected in sequence, a physical information neural network digital twin agent modeling module, a dynamic Pareto front low-latency multi-objective optimization module, and an adaptive decision-making and closed-loop control module. The multi-source heterogeneous data synchronous acquisition module is configured to synchronously acquire cutting force signals, spindle power signals, coolant pressure signals and drilling depth signals at a sampling rate of not less than 10kHz, and complete the timestamp alignment of multi-source data to output structured machining status data. The physical information neural network digital twin agent modeling module is configured to use spindle speed, feed rate, high-pressure internal cooling pressure and current drilling depth as input vectors, and tool wear, cutting temperature and surface roughness as output vectors. The network training is completed by using a composite loss function that integrates data loss terms and physical residual terms. The physical residual terms are embedded with the Oxley cutting mechanics analytical equation and the one-dimensional transient heat conduction Fourier partial differential equation. The dynamic Pareto front low-latency multi-objective optimization module is configured to construct a three-dimensional objective optimization space by maximizing material removal rate, minimizing surface roughness, and minimizing tool wear. Based on the output of the digital twin surrogate model as the fitness evaluation function, the improved non-dominated sorting genetic algorithm NSGA-Ⅲ is used to dynamically solve the Pareto optimal solution set. The adaptive decision-making and closed-loop control module is configured to select the optimal combination of process parameters from the Pareto optimal solution set and send the optimal process parameters to the CNC system through the communication interface to realize real-time closed-loop adjustment of the deep hole drilling process.

2. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The composite loss function is expressed as: Loss = α・Loss_data + β・Loss_physics, where Loss_data is the mean square error between the network prediction and the measured value, Loss_physics is the physical residual term, and α and β are preset weighting coefficients.

3. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 2, characterized in that: The physical information neural network digital twin agent modeling module is also configured to: calculate the partial derivative of the cutting temperature output by the network with respect to the drilling depth, and apply an exponential penalty factor to the prediction results that violate the high-pressure fluid convection heat transfer boundary conditions, and include the penalty results in the physical residual term.

4. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The multi-source heterogeneous data synchronous acquisition module includes a piezoelectric force gauge, a high-frequency current transformer, a coolant pressure sensor, and a CNC pose analysis unit. The output structured machining status data is a standardized JSON object, which includes timestamp, drilling depth, three-dimensional cutting force, spindle power, and coolant pressure fields.

5. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The dynamic Pareto front low-latency multi-objective optimization module integrates Chebyshev scalarization mechanism and reference point guidance strategy, with a single Pareto front solution computation delay of no more than 500ms.

6. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The dynamic Pareto front low-latency multi-objective optimization module is configured to access the current drilling depth and structured processing status data in real time during the evolutionary iteration process, dynamically update the fitness evaluation, and generate the optimal parameter solution set for the next processing cycle.

7. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The adaptive decision-making and closed-loop control module is configured to calculate the Euclidean distance between each candidate solution in the Pareto optimal solution set and the ideal point in the normalized target space, and select the solution with the smallest distance as the optimal combination of process parameters for final execution.

8. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The adaptive decision-making and closed-loop control module uses the Ethernet TCP / IP protocol to write the optimal process parameters into the macro variable register of the CNC system, thereby achieving real-time interpolation coverage of spindle speed, feed rate, and cooling pressure.

9. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 1, characterized in that: The system adopts a microservice distributed architecture based on Spring Boot, wherein: the multi-source heterogeneous data synchronization and acquisition module is deployed as a microservice on the EdgeXFoundry edge computing framework; the physical information neural network digital twin agent modeling module is deployed on a TensorFlowServing container cluster, providing a high-concurrency inference interface.

10. The multi-objective optimization system for high-temperature alloy deep hole drilling process parameters according to claim 9, characterized in that: The dynamic Pareto front low-latency multi-objective optimization module is configured with Redis in-memory database middleware for high-speed caching of processing state data and parallel computation acceleration of multi-objective evolutionary optimization.

Images