【作者】 李奎；

【导师】 张可；

【作者基本信息】 重庆大学 ， 电子信息（专业学位）， 2024， 硕士

【摘要】 压力仪表作为一种精密测量装置,用于量化两个空间点间的压力差,其广泛应用于复杂工业控制系统。在出厂前,需要对压力仪表进行整机标定,整机标定是压力仪表出厂质检过程的最后一个质检环节。然而传统压力仪表整机标定流程复杂,存在检测周期长、检测效率低等问题。因此,本研究聚焦于压力仪表整机标定过程的质量检测及预测研究,以促进压力仪表整机标定的数字化转型。主要研究内容如下:首先,分析了压力仪表的结构与整机标定流程,以差压压力型和表压压力型为典型对象,主要分析了整机标定这一关键的整机标定环节。在此基础上,选取了12个关键整机标定测试参数,并以实际工程数据为基础构建了整机标定测试数据集,为后续压力仪表质量检测及预测研究奠定了可靠的数据基础。其次,考虑到实际工业场景中压力仪表整机标定过程检测疏漏导致整机标定测试数据不完备,降低了整机标定过程的准确率和可靠度。本文以软测量和深度神经网络为关键技术,设计了基于软测量的压力仪表质量检测框架,并搭建了软测量模型和质量检测模型。此外,分析并设计了基于软测量模型度量的整机标定测试流程优化方法。更进一步设计了基于对抗信息增强的软测量框架,以提升软测量模型的建模能力。实验验证了上述方法的可行性和有效性。进一步,为优化生产计划和资源分配,设定了预测下一批次压力仪表整机标定测试数据和合格率的目标。首先针对现有时序预测方法对压力仪表整机标定测试数据时序特性和关联特性建模不充分问题,分别从时域、频域和时频的角度构建整机标定测试数据时序特征,为实现压力仪表整机标定测试数据的时序特性和关联特性建模提供较为完备的特征基础。然后,设计了整机标定测试数据及检测合格率预测框架,主要包括多模态特征提取模块的设计和多模态特征融合及预测模块设计。最后,在压力仪表整机标定测试数据集上验证了所设计方法的有效性和优越性。最后,以上述研究成果为核心,设计并实现了压力仪表整机标定过程的质量预测系统。经过需求梳理、软硬件架构规划、数据采集传输与存储方案设计等阶段,构建了实时数据监控、压力仪表质量检测、压力仪表质量预测、统计分析等交互界面。优化了生产计划,提升了资源调度效率,强化了全程质量管理,从而推动了其向智能制造的升级。更多还原

【Abstract】 Pressure gauges,functioning as sophisticated measurement instruments to quantify pressure differentials between two spatial points,find extensive application within intricate industrial control systems.Prior to deployment,these gauges undergo comprehensive whole-system calibration to fulfill the precision measurement requirements.Nonetheless,conventional calibration procedures for pressure gauges are intricate,beset with issues such as prolonged inspection cycles and low efficiency.Consequently,this research centers on the quality inspection and predictive analysis of the whole-system calibration process for pressure gauges,thereby facilitating their digital transformation.The principal areas of investigation are outlined as follows:Initially,the structure and calibration procedure of pressure gauges are scrutinized,with differential pressure and gauge pressure types serving as paradigmatic cases,emphasizing the pivotal stage of whole-system calibration.Based on this,twelve key parameters for whole-system calibration tests are selected,and a calibration test dataset is constituted utilizing actual engineering data,furnishing a dependable data foundation for subsequent quality inspection and predictive analyses of pressure gauges.Subsequently,considering omissions in the whole-system calibration process of pressure gauges in real-world industrial contexts or interruptions in data transmission links that lead to incomplete calibration test data,thereby diminishing the accuracy and dependability of the calibration process,this work adopts soft sensing and deep neural networks as core technologies.A framework for quality inspection of pressure gauges grounded in soft sensing is devised,accompanied by the construction of both soft sensing and quality inspection models.Furthermore,an optimization strategy for the whole-system calibration test process,guided by metrics from the soft sensing model,is analyzed and designed.Moreover,a soft sensing framework enhanced with adversarial information augmentation is contrived to elevate the modeling capacity of the soft sensing model.Experimental validation attests to the feasibility and efficacy of these methodologies.Additionally,with the aim of refining production scheduling and resource allocation,the objective of forecasting the next batch’s whole-system calibration test data and pass rates is set.Acknowledging the inadequacy of current time-series prediction methodologies in fully capturing the temporal traits and interdependencies of pressure gauge calibration test data,sequential characteristics from temporal,spectral,and time-frequency standpoints are constructed to provide a comprehensive feature foundation for modeling these properties.Thereafter,a framework for predicting whole-system calibration test data and the pass rate is designed,encompassing the design of multimodal feature extraction modules and the integration and prediction of multimodal features.Lastly,the efficacy and superiority of the devised methods are corroborated via the pressure gauge whole-system calibration test dataset.Finally,capitalizing on the aforementioned research findings,a quality prediction system for the whole-system calibration process of pressure gauges is designed and realized.Through stages encompassing needs analysis,planning of software and hardware architectures,and the conception of data acquisition,transmission,and storage solutions,interactive interfaces for real-time data surveillance,quality inspection,quality prediction,and statistical analysis of pressure gauges are established.This leads to optimized production planning,enhanced resource allocation efficiency,and reinforced end-to-end quality management,thereby propelling the evolution towards intelligent manufacturing.更多还原