The Alexa for coatings labs
Voice-controlled digital assistants have become a part of daily life for many people. In the future, they could also make lab work easier. The first such assistant is now real.
Bearing names such as Siri and Alexa, virtual assistants are now a part of the family in one out of five American homes. Voice-controlled digital assistants promise to provide us help in our daily lives. If everything goes according to Dr. Gaetano Blanda’s wishes, this will only be the beginning. Blanda, who works as head of the coating additives business line at Evonik wants to turn voice-controlled digital assistants from simple helpers for daily life into chemistry experts and use them in a place where a comprehensive amount of specialized knowledge is needed, and a technical language is spoken: the laboratory.
In order to exactly meet the customers’ wishes with respect to colour, gloss, and durability, the experts have to create complex mixtures in the lab which they supplement with the right additives. Thousands of combinations are possible — far more, in fact, than the human brain can handle.
“Usual assistants simply can’t understand our language,” says Dr Oliver Kröhl, Head of the Strategic Business Area Development at Coating Additives at Evonik.
The formulators spend correspondingly much time searching through notes and data sheets. So the new digital assistant Coatino could help users research to adjust ingredients directly in the laboratory. “We talked about new ways in which business could develop,” says Dr. Oliver Kröhl, Head of the Strategic Business Area Development at Coating Additives and the project’s manager. “Innovations are no longer just limited to finished products or processes. Instead, you need to demonstrate your ability to come up with solutions in the form of new services and business models.”
From can to prototype
To find out if a digital assistant could work, the scientists coated an empty can of paint in the company’s colours. They then put it into the laboratory, where they filmed a discussion between a colleague and the can. In the video, the user asked the can about a suitable waterborne anti-foaming agent for a wood coating. The can gave its reply, provided the lab employee with a selection of products, and ordered a sample.
“Back then, the questions were answered by a colleague who stood behind a wall,” says Kröhl. “Although this was rather ad hoc, we wanted to tangibly test our idea with customers and quickly get feedback.” The video was shared with several customers and the team also conducted structured interviews.
This approval encouraged the developers to move into uncharted territory. “We’re experts for paints and coatings, but not for voice-controlled assistants,” says Kröhl. “That’s why we knew that the project might not work. However, we and our customers thought it had such great potential that we were willing to take the risk.
Learning the coatings lanuage
This was no easy task, because conventional voice-recognition systems were unable to handle the specialist vocabulary. “The usual assistants simply can’t understand our language,” says Kröhl. They quickly reach their limits when you ask them about dispersion, rheology or silicone resins, for example, and they can, at best, only supply general information.
“They have to be able to do a lot more in order to formulate a coating,” says Kröhl. “If they don’t know the components’ properties and how they interact, they won’t be any help in the laboratory.”
Thousands of combinations
The various components influence each other’s effects, depending on the mixture. The number of possible combinations is immense. Even if only ten curing agents, ten binders, ten pigments, and ten additives are considered during the development of a coating recipe, these numbers translate into 10,000 possible combinations. “Customers have very precise ideas about the capabilities that a product should have once it’s finished,” says Blanda.
In order to develop a functional voice-controlled assistant for the coatings industry, the researchers at first began to structure all of the available information and feed it into a huge database. In the next step, they made it possible to call up this information using a voice-control function.
Training for global application
For example, if you ask the assistant, “Which additive is suited for printing ink?,” the system obviously has to be able to understand each word. Among other things,
“The assistant can tell me which additive would be best suited for my formulation and my requirements”, explains Dr. Gaetano Blanda, head of the coating additives business line at Evonik.
Coatino had to learn that “additive” designates a certain category of coating components. In the next step, the assistant has to access its data, search through it, create suitable links, and assign the data to a possibly relevant result. To do so, it first breaks down the sequence of sounds into their smallest components and conducts a data search based on characteristic properties.
A special challenge for the assistant is that it must be able to understand not only German nouns in the nominative case but also in other cases. The researchers also want to make sure that the speaker’s dialect or accent won’t hamper the result. The aim is to enable Coatino to understand customers’ pronunciations worldwide. Added to these challenges are the speakers’ different talking speeds and pitches as well as the specific context of a discussion.
“The training process is very nerve-wracking,” says Kröhl. “And after the trial run with our colleague in Shanghai was finally successful, it went wrong with our colleagues in Essen.” For almost two years now, Coatino has been jointly developed and trained by the business line and an external development company from Berlin. The assistant passed its first important development test when the prototype was demonstrated at the European Coatings Show 2019 in Nuremberg in April.
When asked about suitable additives, the assistant not only presents a list of products but also prioritizes them. “Coatino can tell me which additive would be best suited for my formulation and my requirements. It can thus give me well-founded recommendations,” says Blanda. Once a user has found the desired product, he can issue a voice command to order a sample, directly call up the pertinent technical data sheet by e-mail or have a conversation with an expert arranged.
New formulations from the database
The Coatino prototype was ready just in time for the start of the European Coating Show. “We immediately presented it to a select group of our customers,” says Blanda. Instead of a can, the users imparted their wishes to a tablet via a microphone. In 2020 the researchers plan to make Coatino available for the entire coatings industry.
However, there is no end in sight for the system’s further development. “When you use digital assistants, you continually come up with ideas for new features,” says Kröhl. For example, Coatino could conceivably not only supply existing formulations but also suggest its own new mixtures. The scientists could directly test these mixtures in the lab and enhance them for their own use. “Our Coatino might one day really become an artificially intelligent entity,” says Blanda. “But we still have a long, long way to go until then.”
Related content
Innovations: “Clear communication between our technical and business teams”
