AI & chemistry, a match made in heaven
AI and big data have been revolutionizing technologies and applications in all industries. Many an article has been written on the fruitful application of AI in the medical sector and the automation sector. Unfortunately, the same cannot be said for the chemistry sector, even though breakthrough AI technologies are transforming the industry. In this article we discuss how AI & chemistry are a match made in heaven.
The spoils of artificial intelligence (AI) and big data are to be shared across industries and applications. Numerous articles have been written on the application of AI in the medical sector, in which algorithms will shape the future of medicine with better, cheaper and more accurate diagnosis and treatment of diseases. Even more has been written on the groundbreaking impact of AI in the field of automation, ranging from reducing labor intensive jobs such as factory work and transportation. Unfairly, significantly less is written about the marriage between the chemical sector and AI. I use the word ‘unfairly’ because the same breakthrough technologies are about the hit the sector hard. The writer of this perspective is of the humble opinion that the chemical industry is especially prone to the innovative power of AI. In this article we will start our journey by clarifying the key technological enablers of AI in chemistry and discuss their strengths and peculiarities. Next, we identify the core opportunities of AI and Industry 4.0 in chemistry and link provide some concrete examples.
Industry 4.0 has become an umbrella term to describe the possibilities that arise from the synergy between internet of things, artificial intelligence and big data. As umbrella terms risk losing their meaning in generality, we feel that a clarification of the key technological enablers is necessary.
Data is being created all throughout the value chain of the chemical sector. On the one hand, databases of molecule interactions have been collected over the years in the search of new chemical/ pharmaceutical compounds. On the other hand, customer data is gathered in CRM systems. Moreover, data can be found everywhere in between: in real-time production line data, real-time chemical process parameters at the line, yield data, manufacturing/resource schedules, order forecasts, the list goes one. Big data itself is not a technological enabler, but a consequence of digitizing processes and centralizing data from many sources. Data infrastructure is a key technological driver together with digitalization of processes in sales, logistics and forecasting. However, the last few years the rapid digitalization of the physical world has been instrumental in the move towards big data applications. This we call “the internet of things”.
Internet of things (IoT)
Historically, the world could be divided into the digital and the physical. In recent years, however, the gap between these worlds has been closing through the advent of digital sensors popping up everywhere. A pressure gauge at a production line became a digital one, chemical process monitoring has been digitized and centralized on a central server for process control, among other things.
Artificial intelligence (AI)
Transforming data into knowledge and actions is crucial to gain a competitive edge. Neither big data nor IoT are value drivers on their own. In fact, it is quite the contrary: gigantic piles of data instantly freeze anybody who dares to take a look and scares them away. This is where artificial intelligence and data analytics come into play. AI allows to not only make sense of the data, but to uncover information that no human could ever find or put to use on their own, as well. The field of AI is vast and has many sub-domains. If you are interested in a detailed exploration of the field of AI, we refer you to our presentation. Nevertheless, we comprehensively summarize this field by listing what AI systems are better at compared to human users. Here we go!
The four strengths of AI
Learning from a lot of data
Machine learning (ML) systems are especially apt to learn making accurate predictions from vast amounts of data, that would simply be too extensive for a person to wrap their head around. A learning machine continue digesting current and historical data without tire.
Learning from data that comes from many sources at once
ML systems also allow to combine many streams of data from different sources to make accurate predictions, where for humans, it becomes impossible to keep track of all the factors that come into play to make that predictions.
Learning complex relations in data
ML algorithms can find highly complex relations between data and predicted values of interests. These relations would be very difficult, if not impossible, to uncover by humans within any reasonable budget.
Considering many options smartly
Typically, tasks like resource or logistics planning and optimization rely on considering many possible solutions under certain constraints. AI systems allow to automate this process and consider more elaborate proposals.
AI & chemistry, a match made in heaven
Now that we know the strengths of AI and have laid out the key enablers in Industry 4.0, we are finally equipped to explore why AI and chemistry are a match made in heaven. We have extracted the key points of the value chain where AI can innovate and have provided some examples of applications.
AI can drive chemical innovation and R&D
Our first stop in the value chain is at the level of chemical innovation and R&D. Historically, research was carried out based on first principles and empirical evidence. This empirical evidence was built on the backs of laboratory technicians performing many experiments, of which the sheer volume is a huge cost driver. AI provides another option to do research. By means of data driven developments the strengths of AI can leverage the huge amounts of historical experiment data. For example, we could consider a computer-aided drug design algorithm to train the ML algorithm, in which we use 743,336 compounds, approximately 13 million chemical datapoints. The model could then provide a classification of properties, structures and functions.
Medicinal chemistry and pharmaceutical research is being revolutionized by data & AI. However, drug discovery and development pipelines are long, complex and depend on numerous factors. ML approaches provide a set of tools that can improve discovery and decision making for well-specified questions with abundant, high-quality data. Opportunities to apply ML occur in all stages of drug discovery. Examples include target validation, virtual screening (VS) for rational drug/chemical development, identifying prognostic biomarkers and analyzing digital pathology data in clinical trials. Applying ML can promote data-driven decision making and could speed up the process and reduce failure rates in drug discovery and development. Other applications that have proven their effectiveness are prediction of biological affinity, pharmaceutical and toxicological studies, as well as quantitative structure-activity relationship (QSAR) models.
Theoretical and computational chemistry are being pushed forward by AI. This allows for predicting chemical, physical or mechanical properties of polymers, ceramics and other materials properties without conducting new experiments.
In analytical chemistry we find applications like neural networks that can calibrate and analyze spectroscopy data, as well as deep learning to yield a tool for both protein design and structure prediction. Neural networks are also being employed in biochemistry and correlated research fields such as protein, DNA/RNA and molecular biology sciences quite successfully.
- Compound classification
Chemists spend hundreds of hours to manually classify compound. Furthermore, manual compound classification is prone to human error and thereby not cost-effective. ML applications are being used for classifying compounds. ML methods are widely popular to conduct cost effective R&D activities such as determining in vivo toxicity. Based on available in vitro data, ML applications can be designed to predict the toxicity of chemicals. ML algorithms can predict the atomization energies of organic molecules. (Von Lilienfeld)
- Chemical reaction prediction
Chemists at Princeton University and Spencer Dreher of Merck Research Laboratories harness artificial intelligence to predict the future of chemical reactions. They predict yields accurately while varying up to four reaction components by applying ML. Researchers were also able to feed chemical components into neural network trained on a dataset of 395,496 reactions. The neural network then used its findings about prior reactions to predict possible occurrences under new conditions.
Implementing AI in chemical engineering & manufacturing
The next stop in the value chain is the field of chemical engineering and manufacturing. Modern manufacturing plants generate huge amounts of data at the production line and in the plant. This data allows leveraging the 4 strengths of AI to tackle old challenges.
Predictive maintenance and fleet based management for machinery. Similar to other industries, the chemical industry is faced with large plants filled with assets that need maintenance. Today this is often done on a scheduled basis, whilst it actually would be more cost effective to detect necessary maintenance or equipment breakdown early and precisely. This would ensure that prevention and maintenance can happen in a targeted way. Predictive maintenance implies using anomaly detection, pattern recognition and other AI techniques to predict when machinery needs maintenance. This is already being applied in factories, fleet management, process control. The models can absorb huge amounts of data, consider specific assets or processes and trigger any abnormal behavior early.
AI in chemical process modeling and optimization of chemical processes can lead to yield increase and waste minimization.
AI chemical process control, monitoring and fault detection reduces cost, prevents human error, optimizes operation and automated quality control. Online automated condition monitoring and control has become a significant tool in ensuring stable operation by alerting deviations from normal operations early. Some examples are high speed and accuracy thermal control of a continuous flow chemical reactor with computer vision and a predictive artificial neural network, online reactor monitoring with neural networks, and deep learning for pyrolysis reactor monitoring in which thermal imaging is used to perform smart monitoring to detect faults using neural networks. Other applications can be found in manufacturing planning and configuration.
Big data will enable smart supply chains, logistics and resource management
This part of the value chain can largely benefit from the strengths of AI. Although planning software has existed for many decades, the new availability of data, combined with increased computation power allow for better application and more accurate algorithms. To exemplify, for many chemical industries, the demand for products fluctuates throughout the year, such as their demand for oil keeps changing every month. What makes it even more complicated, is that demand planning processes can be inaccurate and therefore, too expensive. Nevertheless, demand planning is a critical process that required in manufacturing. ML algorithms are accurate and well-suited for predicting demand. This is similar for order or demand forecasting and (human/material) resource forecasting.
Planning, forecasting and resource management could greatly take advantage from AI’s ability to consider many possibilities to determine the most optimal solution for production planning, stock management, inventory optimization/management, energy consumption optimization and supply chain management. An example of this is the AI tool that Verhaert has built for a semi-finished product manufacturer with multiple plants in multiple countries. The tool examines multiple forecasts, multiple storage facilities, multiple quarters, varying raw material prices, among other things, to determine the production and stocking schedule for a full year. This tool saves several full-time equivalents of planning effort and reduces the cost north of €400.000/year.
Big data enables smart marketing, sales and customer service
Our final stop in the value chain is at the level of smart marketing, sales and customer service. We provide some examples of AI application.
Sales managers face the daunting challenge of trying to predict where their team’s total sales numbers will fall each quarter. Using an AI algorithm, managers can now predict the next quarter’s revenue with a high degree of accuracy.
- AI powered personalized marketing or experience by personalizing customer content
- Predictive recommendations using customer data to recommend products
- Optimizing the selling process for representatives by analyzing client opportunities to create guidance to help close deals
- Help direct sales offer by finding patterns in customer relationship management that have high chance of yielding new business
- Chatbots for better and/or direct customer service
Having read this article, we hope you are inspired to become a leader in adopting AI in the chemical sector. Verhaert can help you develop and deploy AI tools, as well as create a road map to introduce AI into your company.
Download the perspective