Topic :  Industry 4.0 : What it means for the Manufacturing Industries and Sustainability  (Plenary Lecture- I)

Biography :  Dr. Lakshminarayanan Samavedham is an Associate Professor at the Department of Chemical and Biomolecular Engineering, National University of Singapore. Prior to that, he was a Principal Consultant for Mitsubishi Chemical Corporation, Japan. He has worked in the area of Advanced Process Control, Process Data Analytics and Machine Learning for Medical Applications for the past several years. His primary interests relate to the development and implementation of early warning systems, automated diagnosis tools and performance recovery strategies. As Director of the Institute for Applied Learning Sciences and Education Technology (ALSET) at NUS, he is leading the development of intelligent learning assistants that will personalize and deepen the learning experience for students. Lakshminarayanan has earned his Ph.D degree in Process Control from University of Alberta, Canada, Masters degree from IIT Madras and Bachelors degree in Chemical Engineering from BITS, Pilani.

Abstract  : Over the last few years, Industry 4.0 (i4.0) has attracted attention from all industrial sectors around the world. With the internet moving into the manufacturing industries, the evolution of digital technologies has put companies on the verge of major changes in paradigms and approaches to management of processes and equipment at all levels of the supply chain. The theoretical lens of supply chain innovation (SCI) will be used to investigate the implications of i4.0 on supply chain management for both established companies and startups. This talk will also attempt to contribute to the understanding of how i4.0 can be related to the chemical and related industries by contextualizing the concept, and function as a “door-opener” for further research. Potentials, sustainability aspects and some concrete examples will be employed to convey the message. Both positive and negative impacts expected out of i4.0 activities for the businesses as well as the environment will be highlighted.

Topic :  Whither Artificial Intelligence: Artificial or Augmented Intelligence? Are we doing enough?  

(Plenary Lecture- II)

Biography : Dr. Ravindra D. Gudi is a Professor & Head at Department of Chemical Engineering, Indian Institute of Technology, Bombay. He also holds the Institute Chair Position in Artifical Intelligence and Machine Learning. He earned his B-Tech and M-Tech degrees from IIT Bombay and a PhD from the University of Alberta (1995). His research interests lie broadly in process systems engineering & green engineering, i.e. modeling, optimization, control and fault diagnosis of process systems, sustainability in industrial practices.

 Dr. Gudi has served as a Visiting Professor at the Department of Chemical Engineering, University of Alberta, Canada (1997), Department of Chemical Engineering, University of Wisconsin- Madison (2003-04).  Dr. Gudi has published over 135 scopusindexed  papers  and has 9 US patents to his credit,  in various areas of process systems engineering.

Dr. Gudi is a recipient of several awards including the Canadian Commonwealth Fellowship by the Government of Canada (1991-1995), Lovraj Kumar Memorial Award for promotion of Industry Academia Interaction, (July1998 - January 1999) Manudhane Applied Research Award (2006), Herdillia Award for Excellence in Basic Chemical Engineering (2009). He is an Associate Editor of the IFAC journal of Process Control since 2010 and a Guest Editor for Control Engineering Practice. He serves on several technical committees of IFAC. He has been an active consultant to the industry in India and abroad. He is also currently the President of the Automatic Control and Dynamic Optimization Society of India.

Abstract  : AI & ML based approaches to reconstruction & prediction, with a view to assist decision making, have re-emerged with a stronger potential and application spectrum. These approaches have relied quite heavily on the power of modelling tools, both statistical as well as other advanced AI based tools.

 However, in any decision making, there are considerations related to credibility and completeness of the information sources, which can additionally be brought in, to help in the AI and ML based modelling approaches.

 This talk will explore such possibilities and pose (perhaps awkward) questions on the adequacy of AI and ML approaches as they exist today. The talk will motivate alternate complementing modelling paradigms that may help to generate a relatively accurate reconstruction and prediction with a view to improved decision making. Illustrative examples from industrial as well as academic literature will seek to reinforce some of the proposed ideas.

Topic :  Data to Decision Intelligence: Role of Artificial Intelligence and Machine Learning  in Process Industry

Biography : Srinivasa Rao Koyyalamudi (KS) has 26+ years of progressive experience, covering diverse business functions, industry verticals, and geographies.  Currently, KS is working on the incorporation of a new venture based in India. In his previous role at AVEVA, KS was responsible for Refinery Operations Management and Smart Water business globally based in Singapore. Prior to this role, he was heading the delivery and support function of the Industry Solutions business within AVEVA/Schneider Software. KS has a graduate degree in chemical engineering, a postgraduate degree in chemical plant design, an executive program in global business management, and is a certified project management professional (PMP).

Abstract : Data is abundant in process Industry, as we collect humongous data from operations with highest possible resolution /granularity and archive them for longer period. Such data not only provide the view of current state but also provide insights about how the process was reacting to various discrepancies. Key to optimization of any process is the actions performed in understanding such data by extracting model features/correlations and in some cases the relationships are difficult to understand or to extract. Predictive Analytics play a major role in such cases by predicting future events helping decision support.

Application of Decision intelligence to data (information), is not only to provide insights or predict future events (Predictive Analytics) but to provide prescriptive actions (Prescriptive Analytics) that helps in decision support for reacting to a given scenario.

Topic :  Design Thinking Driven Machine Learning Applications in the Internet of Things World

Biography : Dr. Kanchi Lakshmi Kiran (Kiran) has received his Bachelor of Engineering (B.E (Hons)) degree in Chemical Engineering at National Institute of Technology, Durgapur, India and Doctor of Philosophy (PhD) degree in Chemical and Biomolecular engineering at National University of Singapore, Singapore.

 Kiran is currently working as a Vice President at Regional Business Analytics, Consumer Banking, DBS Singapore, where his focus is on the development and deployment of Artificial Intelligence/Machine learning applications and AI governance framework. 

Prior to this, Kiran has worked at McLaren Applied Technologies Pte Ltd, Singapore for three and half years as a Lead/Senior data scientist and at Yokogawa, Singapore for five years as a Senior Research Engineer. Broadly, Kiran is fascinated about the field of process systems engineering (PSE) and its applications in providing practical solutions to multidisciplinary and real-world challenges. Overall, Kiran has 12 years of research and industrial experience in the real-time application of big data analytics, mathematical modeling, controland optimization in diverse fields (Healthcare, Energy, Transport, Smart Manufacturing, Industrial Internet of Things (IIoT)), 8 years of experience in executing proof of concept projects and in providing customer support and 6 years of experience in software development/project management.

Furthermore, Kiran’s scientific contributions have been well recognized across the globe in the form of international patents (4), journal and conference publications, government grants. One of his research proposals related to “improving human reliability, plant reliability and availability in power plants in Singapore” has been awarded National Research Foundation (NRF) grant by the Energy Market Authority (EMA), Singapore in 2014. Based on his contributions as a Chemical Engineer, he was nominated for the Young Chemical Engineer of the Year (2014) Award, IChemE, Singapore.

Abstract : In the recent years, there has been tremendous evolution in gamut of technologies (smart sensors, wireless communication, mobile, embedded systems, data storage, cloud computing) and the devices based on these technologies are well-connected and integrated which is known as Internet of Things (IoT). As a result, colossal amount of data (“The new oil”) is generated every day and it is growing at a surprisingly faster rate.

Currently, in this IoT world, there has been a significant organizational top-down drive to adopt digital transformation path and identify novel opportunities to create revenue from data across all public and private sectors. Consequently, everyone has launched the digital drive and artificial intelligence (AI)/deep learning (DL)/ machine learning (ML) has been embraced as a powerhouse to unlock valuable information from data.  However, there are practical challenges in employing AI/DL/ML based solutions:

• Top management expects estimated Return on Investment (ROI) value by using AI/DL/ML before experimenting it on the real-world business problems.

• Shortage of skill set in setting up the infrastructure for the digital transformation.

• Communication gap between the multidisciplinary teams: Data science experts mostly spend time on developing innovative methodologies with less or no interaction with domain experts from both technical and business teams. Further, some of the domain experts assume that their job is at stake.

• Data silos: Availability of more data for functional level applications but less data for enterprise level applications

• New policies by regulatory authorities related to data usage ethics, data privacy and security and model-based risk assessment.

In this regard, design thinking can play a vital role in implementing AI/DL/ML based solutions. In the presentation, the focus will be on exhibiting the case studies powered by design thinking in industrial automation, healthcare and finance domains.

Topic :  Machine Learning for Learned Machines :: Possibility and Reality of AI and Analytics in Manufacturing

Biography : Dr. Raghuraj Rao is a Chemical Engineer by qualification (B.E : National Institute of Technology Karnataka Surathkal, M.Tech : Indian Institute of Technology Bombay, Ph.D : National University of Singapore), with specialisation in Process Data Analytics, Process Modelling and Optimisation.  He has overall, more than 25 Years of academic, research, plant commissioning, process optimisation and entrepreneurial experience. He is theFounder Director and Managing Director of AKXA Tech Pvt Ltd., a tech StartUp company recognized by Govt. of India. AKXA Tech develops and delivers unique Algorithm based Data analytics Tools/Services to manufacturing sector enabling plant team to take quicker/targeted decisions for enhancing plant stability, productivity and energy efficiency. 

 Abstract : With increasing choices for customers, combined with boundryless competition and price wars, industries are seriously focussing inwards on operational excellence. The large scale manufacturing setups, including process plants, are refocussing on enhancing productivty, plant energy efficiency, higher asset utilisation and reducing time from mine to market. Information Revolution of 1990s has catapaulted, in 2020, into a next generation Industrial Revolution (Industry 4.0).  Artificial Intelegence, Big Data Analytics and Industrial Internet of Things (IIoT) are projected to take manufacturing sector into higher level of productivty and efficiency.  “Machine Learning“  techniques and algorithms are now being harnessed to enhance the performance (seeking Global Optimum) of highly automated and locally optimized equipment (“Learned Machines“). 

But, are the legacy plants with resistance for change,Varyng process conditions (product grades, input quality), Aging Machinery with Learned set of SoPs, ready for this hi-tech migration? Are the envisaged benefits possible across the board/sections of plant?  While the operational team at plant level and the key decision makers are used to mere spread sheet analysis, what is the feasibility of implementing academically advanced Data driven process Modelling, Pattern Recognition, Fault Diagnosis, Predictions and Optimisation techniques and off the shelf generic ML tools to resolve typical plant level pain points? This presentation attempts to bring all these challenging areas and key issues into perspective and takes a birds eye view on Possibility and Feasibility of benefiting from ML with case studies from different manufacturing sectors (Process Plants/Foundary/Automobile/Pharma/Transport).

Topic :  Transitioning from Supervised Machine Learning to Unsupervised Machine Learning for Process Control and Optimization in Chemical Plants

Biography : Srini Palavajjhala received his B.E. (Chem.) from Karnataka Regional Engineering College in 1989, Masters in Chemical Engineering from Indian Institute of Technology, Bombay, and a D.Sc. from Washington University in St. Louis.  Prof. Sundarmoorthy was his final year project advisor and was instrumental in his selecting System Sciences as a field of study in college and later pursuing as a career.

Srini worked at CAD Center at IIT Bombay before starting his doctoral studies.  After receiving his doctoral degree, Srini joined Dynamic Matrix Control Corporation in Houston where he got to work with Charles Cutler.  DMC Corporation was acquired by Aspen Technology in 1996.  Srini continued working at Aspen till 1999.  Srini joined as Vice-President (Technology) at Bass Rock Consulting in Houston and then founded Singular Control Technologies in 2011.

Srini has over 30 years of experience implementing real-time control and optimization solutions at oil refineries, chemical factories, mid-stream gas plants and solar power plants.  He has implemented over 200 real-time control and optimization applications.  Srini has extensive experience in capturing non-linear and non-ideal behavior of chemical processes and reactors in models by designing experiments and generating real-time data.  Srini also works in the area of equipment start-up and troubleshooting instruments and process equipment.

Srini is currently a System Integrator of Aspen APC and GDOT products; an Implementation Partner for Aveva APC, Seeq, and PACE technology; a Consultant to Imubit and an Advisor to Geminus.AI.  Srini provides investment advice through Gerson Lehrman Group and Guidepoint.  Linked In Profile Link - (9) Srinivas Palavajjhala | LinkedIn

 Abstract : “Dynamic Matrix Control (DMC) – A Computer Control Algorithm” by Cutler and Ramaker was published in 1980.  DMC is a supervised machine learning program that delivers value by optimizing and controlling a plant at multiple constraints simultaneously.  DMC uses a dynamic model to predict future values of variables, a linear programming optimizer and a quadratic error minimizing control algorithm.  The linear programming optimizer originally proposed by Cutler and Ramaker was replaced with a nested linear programming solver in the mid-nineties.  Today, DMC uses a non-linear solver and allows non-linear output variables.  DMC models are developed using multivariable closed-loop step testing.  The models are developed using plant data gathered during step testing.  These models are valid over a narrow operating region and require customization to be accurate to predict non-linear variables over a wide operating region.

Real-Time Optimization (RTO) using open-equation, high fidelity first-principles models has found success in ethylene optimization.  Though RTO has been successful for ethylene optimization, successful RTO applications in oil refineries are few.  RTO targets are written down to DMC to create a non-linear optimization and control platform.  RTO are expensive to build and maintain.  Kinetic models are difficult to tune and update.  Consistent and complete data sets to tune these models is time consuming and expensive to gather.  Non-ideal equipment and process behavior require customization.  Solving RTO problems online often results in convergence issues.  Maintaining consistency of models and solutions between RTO and DMC is a challenge.

Lower fidelity, non-linear surrogate models have found more success than RTO in high-value added non-linear refinery optimization applications like Diesel Pool Optimization and Naphtha Pool Optimization.  Optimizers using surrogate models are quicker to develop and deploy.  They are easier to maintain and deliver value by optimizing non-linear variables.  Developing the non-linear equations used in these applications is still custom and are not suitable to develop using plant data.

Data driven supervised machine learning applications using deep learning networks have found limited success in refinery control and optimization.  Some of the challenges facing the deep learning applications deployed are poor data quality gathered from plants;deep-learning applications running with no real-time feedback; difficulty integrating deep-learning applications with multivariable control;and the amount of time it takes to generate data used to updatethese models.  Lack of transparency and maintainability of deep-learning models for control and optimization applications have made it slow for this technology to penetrate the market. 

Recent developments in machine learning algorithms and platforms are making development of high-fidelity surrogate models using digital twin and process data possible.  Implementing surrogate models developed using high-fidelity models for key constraints and decisions variables may result in faster, easier and less expensive solution deployment. Fusing predictions from ensemble of available model sources of varying accuracy into a single predictive model with different fidelity and accuracy is another area that offers a promising future for non-linear optimization and control.  Technology developments in this area are recent and platforms for industrial use are still in the early stages.

Process Industry has many successful applications using supervised learning, linear, dynamic model-based control and optimization.  ROI of these projects are good with paybacks in a few months.  However, the field is still open to build non-linear, supervised machine learning control and optimization applications at oil refineries.  

It is a promising future with new technology in AI and machine learning technologies to imagine models developed using unsupervised learning, updated using real-time plant data and self-tuning optimization and control programs that works over wide operating regions with little to no maintenance required.