Search

CN-121997486-A - Nasal catheter design method based on reinforcement learning

CN121997486ACN 121997486 ACN121997486 ACN 121997486ACN-121997486-A

Abstract

The invention provides a nasal catheter design method based on reinforcement learning, which comprises the steps of collecting oxygen supply working conditions, nasal cavity geometry, material characteristics and manufacturing process boundaries, establishing and parameterizing geometry, material and process data, establishing a simulation environment of coupling fluid, heat and contact pressure based on the parameter space, establishing a simulation prediction model and performing error calibration, determining target performance indexes of the nasal catheter, establishing a composite rewarding function containing leakage rate, contact pressure peak value and size deviation, introducing manufacturing feasible domain constraint, taking the simulation prediction model as the environment, performing strategy training on a parameter set by adopting reinforcement learning algorithm, and mapping optimal geometry, material and process parameters into CAD (computer aided design) mould model, material formula and numerical control equipment control instruction respectively, so as to realize manufacturing floor. The invention realizes the automatic design and manufacture of the nasal catheter under the personalized working condition, and improves the tightness, comfort and dimensional stability.

Inventors

  • CHEN ZEFENG

Assignees

  • 环玺医疗用品(苏州)有限责任公司

Dates

Publication Date
20260508
Application Date
20251231

Claims (10)

  1. 1. The nasal catheter design method based on reinforcement learning is characterized by comprising the following steps of: S1, collecting oxygen supply working conditions, nasal cavity geometry, material characteristics and manufacturing process boundaries, establishing a geometric parameter set, a material parameter set, a manufacturing process parameter set and a quality detection characteristic set, and uniformly parameterizing to form a manufacturing feasible region; S2, constructing a simulation environment of coupling of fluid, heat and contact pressure based on the parameter space, taking the parameter set as input and the quality characteristic as output, building a simulation prediction model and carrying out experimental calibration; S3, determining target performance indexes, constructing a reward function with leakage rate, contact pressure peak value and size deviation as cores, and combining manufacturing feasible areas to form a composite reward system; s4, taking the simulation prediction model as an environment, performing strategy training on the parameter set by adopting a reinforcement learning algorithm, guiding search by using a reward function and constraint, and outputting an optimal design strategy; And S5, outputting optimal geometric, material and manufacturing process parameters according to the optimal design strategy, and mapping the optimal geometric, material and manufacturing process parameters into CAD (computer aided design) mould models, material formulas and equipment control instructions respectively to realize manufacturing and landing.
  2. 2. The method for designing nasal catheter based on reinforcement learning according to claim 1, further comprising S6, performing on-line detection on the manufacturing process, calibrating the process parameters when the measured and predicted deviation exceeds the threshold, and recharging the data with the simulation model and reinforcement learning agent to realize closed-loop updating of design-manufacture-calibration.
  3. 3. The method for designing a nasal catheter based on reinforcement learning according to claim 1, wherein S1 specifically comprises: Collecting target oxygen supply working condition parameters including oxygen supply flow, oxygen supply pressure, nasal cavity geometric constraint, wearing duration and environmental humidity temperature range; obtaining a geometric parameter set required by nasal catheter design, wherein the geometric parameter set comprises a nasal prong angle, an end curvature, a side hole opening ratio, an inner cavity section size and an appearance profile; Acquiring a material parameter set used for the nasal catheter, wherein the material parameter set comprises a material hardness gradient, an elastic modulus gradient, a surface friction coefficient and deformation recovery characteristics; And obtaining a manufacturing process parameter set comprising an extrusion temperature, a traction speed, a cooling curve, a die compensation coefficient and a secondary shaping curve.
  4. 4. The method for designing nasal catheter based on reinforcement learning according to claim 3, wherein S1 further comprises obtaining quality detection feature sets including leakage rate, peak contact pressure, dimensional deviation and consistency index between batches, and performing unified parameterization processing on the geometric parameter set, the material parameter set, the manufacturing process parameter set and the quality detection feature sets to form unified parameter space and manufacturing feasible region constraint.
  5. 5. The method for designing a nasal catheter based on reinforcement learning according to claim 4, wherein S2 specifically comprises: Based on the unified parameter space, a multi-physical field coupling model is established, and the gas flow, heat conduction and nose wing contact pressure distribution process is covered; Taking a geometric parameter set, a material parameter set and a manufacturing process parameter set as input variables, and taking a quality detection feature set as an output target; And establishing a simulation prediction model for rapidly predicting the quality detection feature set under the parameter combination change.
  6. 6. The reinforcement learning-based nasal catheter design method of claim 5, wherein S2 further comprises calibrating the simulation prediction model through experimental data to ensure that the deviation between the prediction result and the actual manufacturing condition is controlled within a set threshold; And taking the calibrated simulation prediction model as an environment model of the reinforcement learning intelligent agent, and realizing closed-loop mapping of parameter change and quality characteristic prediction.
  7. 7. The method for designing a nasal catheter based on reinforcement learning according to claim 6, wherein S3 specifically comprises: Establishing a reward function according to the quality detection feature set, and taking the reduction of the leakage rate, the reduction of the contact pressure peak value and the improvement of the dimensional stability as forward excitation; Taking the manufacturing feasible region constraint as a constraint condition, and applying negative punishment to the action violating the processing or assembly limit; and synchronously binding the rewards, the constraint strategies and the simulation prediction model, and providing a clear optimization direction for reinforcement learning training.
  8. 8. The method for designing a nasal catheter based on reinforcement learning according to claim 7, wherein S4 specifically comprises: Constructing an enhanced learning agent, and defining a state as a historical parameter combination and a quality detection characteristic corresponding to prediction; defining an action as a joint adjustment to a set of geometric parameters, a set of material parameters, and a set of manufacturing process parameters; Adopting a reinforcement learning algorithm based on strategy gradient or near-end strategy optimization to perform continuous action space searching; the intelligent agent performs iterative optimization according to the reward function and the constraint strategy to form a convergence process gradually approaching to the optimal design; And obtaining a design strategy converged under the target oxygen supply working condition, and reasoning an optimal parameter set.
  9. 9. The method for designing a nasal catheter based on reinforcement learning according to claim 8, wherein S5 specifically comprises: reasoning the target oxygen supply working condition parameters by utilizing the design strategy completed by training, and outputting an optimal geometric parameter set, an optimal material parameter set and an optimal manufacturing process parameter set; mapping the optimal geometric parameter set into the compensation sizes of the nasal catheter mould and the die orifice, and forming a specific processing drawing; mapping the optimal manufacturing process parameter set into a complete process path and a numerical control instruction of extrusion, cooling, traction and secondary shaping; And synchronously outputting the mapped parameter results to a production control system to form an integrated manufacturing scheme.
  10. 10. The method for designing a nasal catheter based on reinforcement learning according to claim 9, wherein S6 specifically comprises: based on the manufacturing scheme, trial production or mass production is carried out, and actual measurement data consistent with the quality detection feature set is collected; Comparing the measured data with the output result of the simulation prediction model item by item, and identifying deviation sources of geometric, material and process parameters; performing on-line fine tuning and calibration of continuous parameters of an optimal manufacturing process parameter set according to manufacturing feasibility domain constraints; recharging the calibrated result and the measured data to a simulation prediction model and an enhanced learning intelligent agent to form an incremental training sample; updating the agent strategy to enable the design strategy to adapt to the fluctuation of the actual manufacturing environment and keep the output stability; and solidifying the final geometric parameter set, the final material parameter set and the final manufacturing process parameter set to form a closed-loop system for design, manufacture and calibration.

Description

Nasal catheter design method based on reinforcement learning Technical Field The invention relates to the technical field of intelligent design, in particular to a nasal catheter design method based on reinforcement learning. Background Nasal catheters are commonly used as consumables for clinical ventilation and respiratory assistance, and their geometry, material properties and manufacturing process have a direct impact on ventilation efficiency, wearing comfort and stability in use. The existing nasal catheter design depends on experience parameters or a simple trial-manufacture correction mode, has long design period and insufficient precision, and is difficult to adapt to personalized working conditions such as nasal structures, oxygen supply flow, pressure and the like of different people. In conventional processes, the geometric parameters of nasal catheters are usually obtained by manual mapping or empirical setting, and the material properties and manufacturing process are also determined by manual debugging. This approach is not only inefficient, but it is also difficult to achieve a high accuracy mapping between parameters and performance. Meanwhile, due to the lack of simulation prediction and optimization mechanisms, the stability of the performance of the product is difficult to ensure in actual production. We therefore propose a nasal catheter design method based on reinforcement learning. The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims at overcoming the defects of the prior art, and provides a nose catheter design method based on reinforcement learning, which solves the technical problems in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: a nasal catheter design method based on reinforcement learning comprises the following steps: S1, collecting oxygen supply working conditions, nasal cavity geometry, material characteristics and manufacturing process boundaries, establishing a geometric parameter set, a material parameter set, a manufacturing process parameter set and a quality detection characteristic set, and uniformly parameterizing to form a manufacturing feasible region; S2, constructing a simulation environment of coupling of fluid, heat and contact pressure based on the parameter space, taking the parameter set as input and the quality characteristic as output, building a simulation prediction model and carrying out experimental calibration; S3, determining target performance indexes, constructing a reward function with leakage rate, contact pressure peak value and size deviation as cores, and combining manufacturing feasible areas to form a composite reward system; s4, taking the simulation prediction model as an environment, performing strategy training on the parameter set by adopting a reinforcement learning algorithm, guiding search by using a reward function and constraint, and outputting an optimal design strategy; S5, outputting optimal geometric, material and manufacturing process parameters according to the optimal design strategy, and mapping the optimal geometric, material and manufacturing process parameters into CAD (computer aided design) mould models, material formulas and equipment control instructions respectively to realize manufacturing landing; S6, carrying out online detection on the manufacturing process, calibrating the process parameters when the actual measurement and prediction deviation exceeds a threshold value, and recharging the data with the simulation model and the reinforcement learning agent to realize closed-loop updating of design, manufacture and calibration. S1 specifically comprises: Collecting target oxygen supply working condition parameters including oxygen supply flow, oxygen supply pressure, nasal cavity geometric constraint, wearing duration and environmental humidity temperature range; obtaining a geometric parameter set required by nasal catheter design, wherein the geometric parameter set comprises a nasal prong angle, an end curvature, a side hole opening ratio, an inner cavity section size and an appearance profile; Acquiring a material parameter set used for the nasal catheter, wherein the material parameter set comprises a material hardness gradient, an elastic modulus gradient, a surface friction coefficient and deformation recovery characteristics; Acquiring a manufacturing process parameter set, wherein the manufacturing process parameter set comprises extrusion temperature, traction speed, a cooling curve, a die compensation coefficient and a secondary shaping curve; And carrying out unified parameterization treatment on the geometric parameter set, the material parameter set, the manufacturing