How to deploy sensor data in chemical manufacturing
Jeroen Boeye
Head of Machine Learning
Typically, chemical manufacturers have a vast amount of data collected by the numerous sensors within their production equipment situated over various production sites. Still, a lot of chemical manufacturers leave most of their data and data potential untouched, while the winners in the industry are leveraging the full potential of it.
Artificial Intelligence and Machine Learning can be efficiently deployed on this sensor data to solve the challenges that the chemical industry faces today. In what follows, we highlight four main domains with respective use cases which will help you deriving substantial business value from your untouched sensor data.
Smart energy management
Problem
When it comes to the environmental impact of manufacturing, the chemical industry is mainly blamed. According to Statistics Flanders, the chemical industry remains the largest industrial sector in terms of energy consumption. No wonder that chemical manufacturers feel the increasing pressure from environmental organizations and stakeholders to turn the tide. Being more energy efficient is not only key to reduce costs, but also to become a more sustainable chemical manufacturer.
Solution
In order to select the optimal operating conditions for your chemical production plant, you have to monitor standard variables such as temperature, flows, tank levels, pressures, viscosity, etc. With respect to this, soft sensors can be implemented in your production process to gather these data. How is soft sensor different from a hardware sensor? A soft sensor is an inferential estimator when a hardware sensor is unreliable or unavailable. The soft sensor is used for operator decision support, maintenance, and control purposes. (1) Soft sensors utilize the available on-line measurements, such as the exhaust gas analysis results (e.g., O2, CO2, and other volatile compounds), pH, temperature, pressures, flow rates, stirrer speed. (2)
Further, Artificial Intelligence and neural networks do the job of deriving highly valuable insights from the collected production data. As a result, you are able to set up a more energy-efficient production process to achieve both operational excellence and sustainable manufacturing.
Impact
Since energy is a major cost for chemical manufacturing companies, the results of smart energy management could be huge. A global specialty materials manufacturer, producing a broad range of chemicals and fibres for everyday purposes, states that their data-driven energy program helped the company improve its energy efficiency by 10% with annual cost savings amounting to $30 million.
Аnоther ехamplе is Bоrеаlis, аn Аustriаn сhеmiсаl соmpаnу whiсh is thе wоrld’s еighth lаrgеst prоducеr of pоlyеthуlеnе аnd pоlуprоpуlеnе hаs dеplоyеd аn аrtificiаl intelligеncе prоgrаm tо dеvеlоp dуnаmiс tаrget vаluеs fоr thе еnеrgу cоnsumptiоn оf а plаnt, imprоving thе fаcilitу’s еnеrgу usаgе аnd thus сutting еmissiоns аnd cоsts. (3)
In addition to sharp cost savings, smart energy management is crucial to act proactively on environmental government regulations and to meet the sustainability needs of your stakeholders, which will result in a competitive advantage.
Predictive asset management
Problem
There is no doubt that the chemical industry is a capital-intensive industry. The production processes in a typical chemical production plant heavily rely on the operating machines. As a consequence, unexpected down-time causes big production losses and rising maintenance costs. That’s why chemical manufacturers are eager to figure out how to transform from a reactive maintenance (“the machine shut down and we have to fix it to limit the down-time”) to a proactive maintenance (“we know when it’s time to check the machine to prevent down-time”).
Solution
In order to increase the reliability of your production assets, failures have to be predicted. By installing sensors in the production equipment and aggregating all the data in the data platform, on premise or in the cloud, you are able to gather and monitor essential performance data about your production assets. Subsequently, predictive analytics will tell you when it’s the best time to repair the machines before they fail. As a result, predictive asset management sharply reduces the risk of unexpected down-time and high maintenance costs.
Оnе оf thе Dеlоittе's cliеnts, a glоbаl сhеmiсаls соmpаnу rеpеаtеdlу fасеd unplаnnеd dоwntimе duе tо аn еxtrudеr thаt fаilеd mоrе thаn 90 timеs in оnе уеаr lеаding tо lоssеs in prоductiоn, sсrаp, аnd оvеrtimе lаbоr. Using rеаl-timе mоnitоring, thе соmpаnу gаthеrеd structurеd dаtа frоm thе еxtrudеr sеnsоrs аs well аs unstructurеd dаtа frоm mаintеnаncе rеcоrds, trаining rеcоrds, аnd оthеr sоurcеs, аnd dеvеlоpеd fаilurе prеdictiоn mоdеls. Bу еvаluаting cаusе-аnd-еffеct rеlаtiоnships, thе prеdictiоn mоdеl gеnеrаtеd аlеrts аnd rеcоmmеndаtiоns оn thе еxtrudеr pеrfоrmаncе. (4)
Impact
There are several drivers for chemical manufacturers to opt for predictive asset management. First of all, business results include an 80 per cent reduction in unplanned downtime and operational expenditure savings of about $300,000 per asset. In addition to economic reasons, predictive asset management contributes to the safety of people and processes. Last but not least, the aging workforce is a big issue for numerous amount of chemical manufacturers. Luckily, predictive asset management reduces the amount of technicians needed and thus can solve the problem of retiring technicians in combination with the scarcity on the labor market.
Process management and control
Problem
Process variability results from a variety of factors, starting from the quality of the raw materials to variations in internal processes such as raw material dosing, temperature control, residence times, system fouling and aging catalysts. As a consequence of process variability, chemical manufacturers face the problem of output variability. This obviously leads to a lower batch consistency and quality, which isn’t a good case for meeting the customers’ needs. That’s why chemical manufacturers seek ways to gain more control on their production processes.
Solution
In order to optimize operational parameters and so reducing process variability, advanced process control (APC) is often considered by manufacturers. Nowadays, Artificial Intelligence, Machine Learning and Deep Learning techniques can be deployed to create a reliable APC system in your chemical production plant. Thanks to the data collected by numerous sensors, the APC system can recognize patterns in the data and show you deviations in chemical processes before they occur, thus reducing production risks. By suggesting you to adjust some production parameters (e.g. temperature), the APC system prevents process variability.
Оne of the lеаding semicоnductоr prоductiоn еquipmеnt mаnufаcturеrs chоse АI-PС tо hеlp cоntrоl оnе of its nеxt-gеnеrаtiоn sеmicоnduсtоr fаbricаtiоn tооls. Thеy hаd a lаrgе numbеr оf prосеss pаrаmеtеrs thаt nееdеd tо bе cоntrоllеd with еаch wаfеr аnd tо idеntify thе right pаrаmеtеrs rеquirеd a high-fidеlitу simulаtiоn thаt wаs соmputаtiоnаllу еxpеnsivе and slоw. This limitеd hоw mаnу timеs thеу cоuld updаtе thе prоcеss pаrаmеtеrs. With AI-PC, thеу аrе аblе tо simulаte thоusаnds of pоssiblе futurеs еаch sеcоnd аnd аpplу thе bеst prоcеss pаrаmеtеrs in rеаl timе tо еаch wаfеr prоducеd. (5)
Regardless of the industry or the physical process being modeled, continuously improving feedback, real-time feedforward and predictive control make AI-PC the future of process control.
Impact
The real-time analytics and automated control actions of the APC system result in multiple business benefits for chemical manufacturers. First, the APC system’s predictions, alerts and prescriptive responses lower the need for manual reviews and saves operators’ time and effort. Second, having more control over the production processes leads to less process variability and so more batch consistency and quality. This ensures a greater ability to avoid waste and meet the customers’ needs.
Production simulation
Problem
Optimizing a production process isn’t as easy as it sounds. The operators of a chemical production plant can’t just manipulate some production variables to find out whether their experiments result in a more efficient production process. This could turn out quite badly and cause huge losses. Nevertheless, without experimentation, you can’t find a way to achieve operational excellence and thus save costs. That’s why chemical manufacturers want to conduct experiments in a risk-free way to gain insights from their production processes.
Solution
Thanks to Artificial Intelligence, chemical manufacturers are able to simulate their production processes. This virtual copy of real assets or products is called a digital twin. Based on the data collected by the many sensors in their production plants, an intelligent agent can predict how their production processes will vary if some parameters are adjusted. This allows chemical manufacturers to gain insights from their production processes and using these insights for optimization. A specific example is yield optimization, by which chemical manufacturers can find out how to adjust their production parameters to achieve the optimal production output.
Sinopec Engineering, a Chinese chemicals company, used SmartPlant 3D, an advanced plant design software, to plan the plant structure, machinery, and piping models for a 300,000-ton polyethylene project in Maoming and improve workflow. (6)
Impact
By simulating your production processes, you will obtain a better understanding of the impact of changes in the production parameters. Such insights in your production processes are crucial for optimization, which will result in a sharp reduction of wasted resources. Logically, this will lead to a sharp reduction in production costs and increased productivity.
How to translate the relevant AI use cases to your organization and implement them? Faktion can help you identify which AI use cases apply to both your chemical production processes and organizational strategy.
References
(2) Robert Spann, Anna Eliasson Lantz, Krist V. Gernaey, Garkan Sin,
Modelling for Process Risk Assessment in Industrial Bioprocesses,
Reference Module in Chemistry, Molecular Sciences and Chemical Engineering,
Elsevier, 2018, ISBN 9780124095472,
https://doi.org/10.1016/B978-0-12-409547-2.14356-2.
(3) https://cnr.ncsu.edu/news/2020/03/artificial-intelligence-chemical-industry/
(5) https://tignis.com/2021/03/16/intro-to-ai-pc-the-next-level-of-process-control/
