(282g) Differentiable Physics-Based Surrogate Models and Automated Sensor Placement Optimization for Efficient Risk Control of Chemical Leaks

The accidental release of hazardous chemicals can lead to large-scale industrial disasters, such as fires, explosions, or poisoning. Detection and appropriate response to chemical leaks are crucial in preventing such accidents. In this study, we propose the optimization process defines a refined risk function and objective function that includes coverage, rather than just reducing detection time. This allows for more accurate and effective optimization of sensor placement, specifically tailored to real-world industrial scenarios, while still operating robustly in new scenarios. We present a framework for detecting chemical leaks using a differentiable physics-based surrogate model and a sensor placement optimization tool, which combines a limited number of sensors. The surrogate model is developed based on differentiable physics and provides faster results than traditional CFD tools. By optimizing the sensor placement, this framework can quickly respond to chemical leaks. The surrogate model can also be combined with a CNN-based deep learning model to provide high-resolution outputs from low-resolution inputs. Using MINLP, sensor placement optimization can be performed, taking into account various constraints. Overall, this study has the potential to contribute to the development of effective solutions for detecting and preventing the uncertainty of chemical leaks.