Real-Time Prediction System for Biological Fermentation Parameters Based on an AI Electronic Nose: Universal Software and Hardware Architecture and Verification

Real-time monitoring of key parameters (e.g., substrate) is crucial for the precise control of biological fermentation processes. To address the technical bottlenecks of significant lag in offline analysis and the limitations of traditional online sensors, this study de-signed and implemented a universal AI-enabled electronic nose system. The system features a modular hardware architecture integrating a high-sensitivity MOS gas sensor array, a precision constant-temperature chamber, and low-noise signal acquisition circuits to ensure signal stability. On the software side, a software architecture was designed based on the RUP 4+1 view model, employing multi-threaded technology for parallel data processing. An innovative five-stage sampling period was designed to match the dynamic response of MOS sensors, facilitating reliable data acquisition. Combined with a truncated average filtering strategy and peak response feature ex-traction, a lightweight single-hidden-layer neural network model was constructed for real-time prediction. Taking the real-time prediction of methanol concentration during glucoamylase fermentation by Pichia pastoris as a case study, the system demonstrated outstanding performance: R² reached 0.9998, RMSE was 13.5326 ppm, and the prediction delay was less than 1 second. The proposed system provides a robust, efficient, and universally applicable hardware-software solution, demonstrating significant potential for intelligent biomanufacturing.