Weak Spectral Signal Dynamic Characterization Method For Residual Oxygen Concentration Online Detection In Pharmaceutical Glass Vials

Pharmaceutical glass vials serves as a crucial form for sterile preparations,with its sealing integrity directly impacting the health and safety of users.Leakage will result in excessive oxygen concentration,which will deteriorate the quality of medicine in the vials.Therefore,in-situ online detection of oxygen concentration is vital for ensuring drug quality.The detection process generally adopts two methods:concentration threshold classification or precise concentration inversion.Tunable diode laser absorption spectroscopy based on wavelength modulation spectroscopy（TDLAS/WMS）offers rapid,accurate,and non-contact detection for assessing gas concentration.However,open environments,low oxygen concentration,short optical path,and weak oxygen spectrum absorption will degrade TDLAS/WMS performance and accuracy.To address the issue,this thesis focuses on dynamically characterizing weak spectral signals in pharmaceutical glass vials under the open environment.The contributions and innovations in this thesis include:（1）Aiming at the practical problem of frequent fluctuations of demodulated harmonic signals in the open environment,a dynamic characterization operator of harmonic amplitude dispersion（HAD）is proposed.This operator quantifies the signal confidence and guides the construction of an extreme learning machine classification model based on fast matching of harmonic signals.The smoothing of unstable harmonic signals is realized for the rapid and accurate classification of the residual oxygen in the vials.（2）Aiming at the practical problem that it is difficult to distinguish the harmonic signals of adjacent concentration,a dynamic characterization operator of harmonic amplitude pass probability（HAPP）is proposed.This operator converts traditional aliasing-prone amplitude information into classification information with probability labels by performing normal fitting on the distribution probability map of the neighbor concentration.Combined with the tail-to-tail first-order Bayesian topological classification network,the fuzzy classification of the concentration of neighbors under weak spectral signals is realized.（3）Aiming at the practical problem that the single magnitude feature is easily disturbed by noise,a dynamic characterization operator of harmonic characteristic triangle（HCT）is proposed.From the view of harmonic morphology,the operator mines the potential concentration representation information in the traditional 2^nd harmonic signal,and constructs a harmonic characteristic triangle with several features.Finally,inversion models are constructed by using optimized multi-dimensional geometric features,which improved system’s anti-interference ability.（4）A TDLAS/WMS-based online detection system for residual oxygen concentration detection is developed.The system is built based on Alinx’s ZYNQ7020 FPGA development board with the high-speed 14-bit AD and DA.The wavelength step of laser is as fine as 0.61pm,and the acquisition resolution is as high as 0.122m V.In particular,to achieve long-term stable control of output wavelength,a laser self-diagnosis method based on feedback control strategy is proposed.Then,to accelerate the demodulation process of weak absorption spectrum signal,a novel logic for oxygen absorption spectrum demodulation is designed in FPGA platform.Finally,the system is verified at a real production line under the speed of 200 vials/min,and the test results meet the requirements of the Chinese pharmaceutical machinery industry standard JB/T20198-2020.In order to solve the problem of performance degradation and inaccurate detection of TDLAS/WMS technology in the process of online detection of residual oxygen concentration in pharmaceutical glass vials,a variety of dynamic representation operators are proposed to reveal the potential rules of the open environment and explore the potential characteristics of harmonic signals.Then,these operators are further extended to the corresponding concentration classification and inversion models.Finally,the detection system integrated with the rapid concentration classification model has completed the application test on the real production line,providing theoretical support and method reference for further improving the detection accuracy and stability of the sealing quality for sterile preparations.