A Deep Learning Approach to Soft Sensor Design and Process Optimization for an Industrial Nickel Extraction Process

The objective of this project is to use artificial intelligence (AI) approaches to solve complex industrial problems. The two biggest advantages of AI-based approaches are the ability to continuously learn and also learn adequately from historical data. Traditionally, many process information are unmeasurable during live operations because of instrumentation limitations. Also, plants are not sufficiently optimized to maximize production quality, while minimizing waste. Using AI-based approaches, we can develop complex non-linear models from historical to predict the unmeasurable process information. The models are also continuously learning from the new data coming into the plant. To optimize the process operations, another family of AI algorithms called reinforcement learning will be used. These algorithms will learn the whole process, including what happens when each process variable is changed. With this knowledge, reinforcement learning can then provide the optimal sets of inputs to maximize the plant productivity, while minimizing its waste.

Faculty Supervisor:

Jinfeng Liu

Student:

Rui Nian

Partner:

NTwist

Discipline:

Engineering - chemical / biological

Sector:

Energy

University:

Program: