This paper provides a comprehensive survey of machine learning (ML) techniques and their practical applications in industrial sensing, optimization, and control. It highlights key statistical and ML techniques that have been successfully used in process industries, focusing on soft sensing, process optimization, and control. The authors discuss the challenges and benefits of hybrid modeling, which combines knowledge-driven and data-driven models to overcome their respective limitations. They also explore the use of reinforcement learning (RL) as an emerging paradigm for solving complex control tasks. Key findings include: - **Soft Sensing**: Data-driven models, particularly artificial neural networks (ANNs), have seen significant success in soft sensing due to their ability to capture complex process dynamics. Knowledge-driven models, such as Kalman filters, are also used but require more expert knowledge. - **Optimization and Control**: Hybrid modeling is widely used in offline process optimization and real-time optimization (RTO). Model predictive control (MPC) and RL are emerging techniques that integrate model-based and model-free approaches to enhance control performance. - **Reinforcement Learning**: RL has shown promise in process control, particularly in synthesizing existing control structures like PID controllers and constructing hierarchical control systems. The paper emphasizes the importance of balancing deep learning techniques with domain knowledge to ensure interpretability and efficiency in industrial applications. It also discusses the practical considerations and challenges in implementing these techniques, providing insights for practitioners and researchers.This paper provides a comprehensive survey of machine learning (ML) techniques and their practical applications in industrial sensing, optimization, and control. It highlights key statistical and ML techniques that have been successfully used in process industries, focusing on soft sensing, process optimization, and control. The authors discuss the challenges and benefits of hybrid modeling, which combines knowledge-driven and data-driven models to overcome their respective limitations. They also explore the use of reinforcement learning (RL) as an emerging paradigm for solving complex control tasks. Key findings include: - **Soft Sensing**: Data-driven models, particularly artificial neural networks (ANNs), have seen significant success in soft sensing due to their ability to capture complex process dynamics. Knowledge-driven models, such as Kalman filters, are also used but require more expert knowledge. - **Optimization and Control**: Hybrid modeling is widely used in offline process optimization and real-time optimization (RTO). Model predictive control (MPC) and RL are emerging techniques that integrate model-based and model-free approaches to enhance control performance. - **Reinforcement Learning**: RL has shown promise in process control, particularly in synthesizing existing control structures like PID controllers and constructing hierarchical control systems. The paper emphasizes the importance of balancing deep learning techniques with domain knowledge to ensure interpretability and efficiency in industrial applications. It also discusses the practical considerations and challenges in implementing these techniques, providing insights for practitioners and researchers.