Predicting and preventing production losses with AI

Israeli start-up Seebo’s machine learning technology combines artificial intelligence (AI) with insight into manufacturing processes to enable manufacturers, including those in the chemicals sector, to predict and prevent future production losses.

Co-founded by brothers Liran and Lior Akavia in 2014, Seebo’s tools can reveal hidden inefficiencies in production processes - Seebo is an amalgamation of the English word ‘see’ and the Hebrew word ‘bo’, which means inside.

It has become more crucial than ever for chemical producers and other companies to improve efficiency and respond to changing consumer behaviour – for example shifts in demand caused by Covid-19. ‘Manufacturers who wish to stay competitive can no longer afford to tolerate recurring process inefficiencies,’ stated Yanai Oron, general partner at Vertex Ventures, which led Seebo’s latest funding round in March, and tripled its previous investment in the firm. ‘Preventing these losses has now become a strategic priority,’ he said.

‘Difficult and painful’

Liran, who earned a degree in information engineering and computer science from Technion Institute of Technology in Israel in 2003 and began his career as a software developer, is Seebo’s chief operating officer. Lior, who completed an MBA from Tel Aviv University in 2006 while working as a software developer and R&D manager in an elite cyber-security unit in the Israeli army, is chief executive.

During his mid-20s, Liran lived in China. After Lior joined him there, they co-founded Playfect in 2007, which manufactured mobile device and video game accessories for video game consoles, PC, smartphones and tablets. Mobile Fun, a UK-based online retailer of mobile accessories, acquired Playfect in 2013, and the pair then launched Seebo.

We are embedding manufacturing knowledge into an algorithm, and we build algorithms that can learn the specific manufacturing process

‘At Playfect, we experienced the challenges with manufacturing inefficiencies – it was difficult and painful,’ Liran recounts. ‘We came to understand just how important efficient manufacturing software is, and decided to build technology that understands the manufacturing process.’

The brothers recognised that generic AI analysis – no matter how advanced – simply cannot handle the complexity of chemical manufacturing. They soon realised that process knowledge had to be embedded within the algorithm.

Making chemicals involves complicated processes that produce complex and noisy data. Simply running any AI algorithm on that data won’t take into account all of those intricacies – including, for example, parallel processing in which the process steps are split up so that formulation, blending, packing and cleaning can take place simultaneously, and multi-product lines that produce a range of products.

‘That was the reason we invented a process-based AI,’ Liran tells Chemistry World. ‘Anyone who would like to provide value to chemical manufacturers using data will need to marry AI and chemical manufacturing knowledge – we are embedding manufacturing knowledge into an algorithm, and we build algorithms that can learn the specific manufacturing process.’

 Seebo’s AI uses ‘automated root cause analysis’ to identify patterns in production processes and alert teams about why inefficiencies are happening. The technology also uses ‘predictive recommendations’ to prevent future process inefficiencies, as well as ‘proactive alerts’ to indicate in real time when manufacturers need to act to address a specific problem.

The primary reason these losses occur is because there is a root cause hidden within the complex data that eludes the human eye

The system can bring together all the various and siloed data on a given production line, then build a digital model that replicates the whole production process – from raw materials through to the end product. This means that the algorithms aren’t just looking blindly at the data but actually understanding the unique complexities of a particular process. The technology can then illuminate inefficiencies that lead to losses in key areas like quality, yield and throughput.

Liran says Seebo’s models should help companies better understand their processes and substantially reduce production losses from causes like: undesired side products and impurities; raw material variations; sub-optimal reactions; and fouling – which encompasses anything that grows on an immersed surface that shouldn’t be there.

It is impossible for human beings to manually conduct continuous, multivariate analysis on a complicated chemical production process because there is too much data that is constantly changing in very subtle ways, Liran explains. Even self-service analytics tools can only test existing theories and can’t discover things that they aren’t searching for, he adds.

While Liran acknowledges that there are many inherent inefficiencies in the chemical manufacturing industry, and it is impossible to have a 100% efficient plant with zero production losses, he says many of those losses can be reduced significantly. ‘The primary reason why these losses occur at the scale they do is because there is a root cause hidden within the complex data that eludes the human eye,’ he explains.