Development of AI-enabled Sensors for Real-Time Industrial Chemical Monitoring

Thesis Abstract

The rapid advancement of industrial processes necessitates the development of real-time chemical monitoring systems to enhance safety, efficiency, and environmental compliance. Current sensor technologies are often limited by sensitivities, response times, and the inability to adapt to variable industrial conditions, which hampers the prompt detection of hazardous chemical deviations. This study aims to develop and validate artificial intelligence-enabled sensors capable of providing accurate, real-time monitoring of key industrial chemicals, with a focus on volatile organic compounds (VOCs), sulfur compounds, and heavy metals in chemical manufacturing plants. The specific objectives include designing sensor prototypes integrated with machine learning algorithms, evaluating their performance in laboratory and field settings, and establishing predictive models for chemical concentration fluctuations under various operational conditions. The research adopts a mixed-methods approach, combining experimental sensor development with quantitative data analysis and qualitative assessment of sensor performance and applicability. The population for this study encompasses chemical manufacturing facilities in the region, with a specific focus on sampled operational environments in five plants. A total of 50 sensor modules will be developed and tested through purposive sampling, with data collected from continuous chemical sensing over a 12-month period. Primary data will be gathered via prototype sensors integrated with microelectromechanical systems (MEMS) and powered by energy-efficient embedded systems, complemented by laboratory calibration using gas chromatography-mass spectrometry (GC-MS) as the benchmark analytical technique. Data analysis will employ regression analysis and machine learning models, particularly support vector regression (SVR) and convolutional neural networks (CNN), to establish predictive relationships between sensor outputs and chemical concentrations. Model performance will be evaluated through metrics such as root mean square error (RMSE), coefficient of determination (R²), and receiver operating characteristic (ROC) curves to ensure robustness and accuracy. The sensor system architecture will incorporate the Theory of Sensor Fusion to optimize multisensor data integration, supported by the Technology Acceptance Model (TAM) to gauge usability and deployment potential within industrial settings. Anticipated findings include the demonstration of enhanced sensitivity, faster response times, and improved predictive accuracy of AI-enabled sensors compared to conventional analytical instrumentation. It is expected that the integration of machine learning algorithms will significantly improve the sensors' ability to detect and predict chemical concentration fluctuations amidst complex industrial backgrounds. Additionally, the study aims to establish a comprehensive framework for deploying AI-driven chemical sensors in industrial environments, facilitating proactive hazard management and process optimization. This research contributes to knowledge by pioneering an interdisciplinary approach that combines sensor technology, machine learning, and industrial process control, advancing the frontier of real-time chemical monitoring systems. It extends theoretical understanding by applying the Sensor Fusion Theory and the Diffusion of Innovation Theory to sensor development and adoption contexts. The expected outcome is a scalable, cost-effective AI-enabled sensor platform capable of transforming industrial chemical monitoring, reducing reliance on laboratory-based methods, and promoting safer working conditions and sustainable environmental practices. The study concludes that AI-enabled sensors, when properly designed and implemented, can revolutionize industrial chemical management by providing rapid, accurate, and actionable data streams. Recommendations include the integration of these sensors into existing control systems, policy adjustments to support technological adoption, and further research into sensor miniaturization and multi-chemical detection capabilities. Future studies should explore long-term deployment, system resilience, and the potential for integrating these sensors into Internet of Things (IoT) frameworks to facilitate comprehensive industrial process automation and environmental monitoring.