In an exclusive interview with Industry Outlook, Nitin Sambhare, Managing Director, Desmet Engineering, shares his views on the AI and ML integration in energy-intensive food and biofuel plants, the challenges associated with these integrations, and more. He is a global business leader with over 30 years of experience in leading global teams, heading pan-India teams, and manufacturing operations across various industry verticals & domains.
Integrating AI/ML in industrial plants involves combining data, such as mechanical and process-specific design data with real-time process parameters. These data sets are integrated through automated systems. A majority of these plant design-specific data sets are not accessible to the general public. As a result, there are few restrictions on using Gen AI-based solutions to increase productivity in continuous process plant environments. However, when design and engineering firms manage to combine plant-specific design data sets that are accessible to them through internal automation and AI and ML expertise with the real-time process parameter from the process plant owners, the use of AI/ML technologies in developing production and optimization models for increasing efficiencies is still feasible. Additionally, the majority of food and biofuel producers have expressed interest in its precise prediction and optimization models. Although it is a gradual process, new technology takes time to become established, and manufacturers have achieved advances in implementing it, particularly for its accuracy and applicability for real-time plant operation.
Large pieces of equipment used by plants, such as compressors, pumps, and motors, make it difficult to spot anomalies in equipment functioning. Plant owners schedule maintenance to identify these anomalies earlier to reduce abrupt plant shutdowns, production losses, and additional maintenance costs. Sensors on heavy equipment detect anomalies and improve prediction models. AI technologies develop predictive models based on historical plant data from equipment sensors and real-time process parameters and also detect anomalies and early warning indications of equipment breakdown. By frequently updating these predictive models with new data, machine learning improves its accuracy. Prediction accuracy is enhanced by AI and ML to identify indications of gradual damage or degradation. For plant maintenance, AI and ML require integration with advanced automation systems, instrumentation, and a design process.
The emergence of big data has introduced challenges related to data accessibility, relevance, and security. Relevant datasets are essential for AI/ML applications to be reliable, and accurate and their availability requires adopting the right technologies and the strategic use of appropriate data sources. The challenges in food and bio-feed plants are handling multiple technologies and data sources addressed by designing a robust automation. The plant data resides within the automation system since automation engineers require an understanding of the plant's operational processes to coordinate the automation system with operational needs, minimize the potential for errors, optimize performance, and maintain efficiency throughout the plants.
To store and retrieve data from multiple sources, systems, and protocols require additional hardware as each system needs different hardware. When integrating third-party software, each data source introduces its own third-party software and security risks. Third-party software threatens the overall integrity of the IT or plant network slows down response time and reduces system efficiency.
Edge computing offers a solution for integrating software by minimizing the need to store data in a centralized location. Instead, data is processed closer to its source, significantly reducing data transfer latency. When processing the data locally, it remains within the Level 0 and Level 1 commercial plant automation networks, thereby enhancing security. AI/ML applications are more secure and improve effectiveness which enables faster and more accurate decision making by local data processes. In food and biofuel production, real-time monitoring and rapid responses are crucial to maintain operational efficiency and quality.
Real-time data sources are integrated through an automation system that enables the automation of sensors such as instrumentation, monitoring systems, third-party data, and process parameters. Most existing plants were built with older automation systems designed using the technologies and data sources available at that time. The older systems present challenges in compiling and processing data when advanced data sources are integrated into `them. The varying protocols used by older systems, complicate the integration of modern technologies, such as AI and ML, by limiting data accessibility and availability for advanced applications.
Upgrading to modern automation systems facilitates the integration of AI and ML applications, as well as overall improved plant process control, reduced downtime, and enhanced total cost of ownership. The technological advancements significantly reduce the maintenance costs, the hardware footprint, the adoption of the latest integration protocols, and the requirement for engineering efforts to build or design systems. Many food and biofuel plant operators and owners remain hesitant to migrate from legacy systems to the latest systems. The primary barrier is the need for in-depth knowledge and experience in automation to ensure a seamless, reliable transition to modern systems. Unfortunately, few engineering companies that supply these process plants have internal automation expertise.
The AI/ML algorithms currently installed are insufficient to address future challenges. The complex requirements of food and biofuel products have led to the development of neural networks and quantum computing algorithms in AI/ML. Operator training simulators and process simulators are widely used to train plant operators to manage plant operations efficiently. These simulators are often expensive and not entirely accurate as they are built based on specific data points at the time of their design and are not regularly updated as plant dynamics change over a period of time due to gradual damage or deterioration in the equipment and process hydraulics. AI and ML extend a solution by enabling real-time simulation of the entire plant’s process dynamics and are used to create operator simulators and process simulators that continuously reflect the current state of plant operations, including any changes due to equipment deterioration or process shifts. These simulators can be easily updated or retrained on an ongoing basis, allowing for more accurate, real-time training through the use of AI and ML.
.jpg "Artificial Intelligence to Revolutionize Advanced Chip Design")
