Mass Spectrometer Peak Shape Calibration for Overlapping Gas Peaks

The system performs preliminary optimization of peak shapes using machine learning before analysis. By pre-training the peak shape models on reference data and optimizing instrument parameters in advance, the system prepares the optimal configuration for extracting information from all peaks including minor ones hidden under larger peaks, without needing aggressive de-convolution that would lose information.

The patent replaces the traditional mechanical de-convolution process with a machine learning-based peak shape optimization system. Instead of using mathematical de-convolution algorithms that struggle with overlapping peaks, the system uses trained neural networks to directly predict compositions from optimized peak shapes, thereby preserving information from minor peaks while achieving accurate separation.

The system optimizes multiple instrument parameters simultaneously including RF voltage to DC voltage ratio, emission current, and bias voltages rather than using limited scaling factors. By changing and optimizing these critical parameters using machine learning, the system achieves superior estimation accuracy and sensitivity across the full mass-to-charge ratio range while maintaining automated calibration simplicity.

The system performs automated self-calibration using machine learning algorithms that automatically optimize peak shapes and instrument parameters without requiring manual intervention. The machine learning model self-adjusts the calibration based on reference data, eliminating the need for complex manual calibration procedures while achieving high accuracy across all mass-to-charge ratios.