Numbers don’t lie. Haxio CEO Alexandre Gervais repeated that to himself before meeting with his client. In the conference room, Alexandre explained his discovery. To create an accurate and efficient system with a reasonable budget, there seemed to be only one possible path: artificial intelligence.
“He was certainly surprised when he was shown the results,” he recalls. “Using artificial intelligence in the biomedical field is an idea that's outside of the box. But the finding was there.”
After a few days of thinking it over, the client supported them financially and Haxio could move forward. All that remained was to find a team capable of creating high-level artificial intelligence.
Alexandre Gervais, founder of HAXIO, posing in front of his FactorySight project
Alexandre has been interested in technology and manufacturing for a long time.
“I’m the black sheep with my management path. I remember going with my father to factories to see him working on interfaces that go back 20 years. Honestly, it hurt the eyes.”
Alexandre is not the only one to think that. CIRANO, the inter-university centre for research in organization analysis, recently published the ninth edition of its “Québec économique.” In Canada, 42% of small and medium-sized manufacturers have not yet moved to digital, while their competitors in the United States, Europe and Asia are already deeply engaged. This observation makes experts nervous.
Haxio specializes in industrial vision. It was hired by a client to create a mechanism to efficiently inspect syringes pre-filled with saline solution. Since the syringes will be used by healthcare professionals, care must be taken to ensure that they contain no contaminants, that they are sufficiently full and that they generally do not present other defects.
Until then, this work had been done by employees. Eyes squinted, they watched the syringes parade before them at a pace dictated by the machine. Concentrating on these objects that all looked alike, they had to do their best to find the defects: particles, level too low or too high, cap badly placed. Not only was the risk of error significant, but the task was so demanding that employees had to rotate every 15 minutes to be able to maintain their performance.
Clearly, an industrial vision system could mitigate this shortcoming, but current solutions on the market were so expensive that they were only attractive for syringes filled with very expensive drugs.
So Haxio leaned into the problem to build an efficient and less expensive machine. It quickly became apparent that a traditional approach with machine vision would not produce great advances. Such a system would call for programming "if...then” rules. The program had to be told to evaluate each pixel so that if a pixel were black, according to this or that condition, it must conclude that the syringe is to be eliminated. The pitfall is that you have to imagine all possible situations to write enough rules that allow you to achieve high accuracy.
Source: Optima Packaging Group is a German company that automates syringe filling processes
Artificial intelligence brought a solution. Using deep learning, the machine can be left to learn by itself to recognize imperfections from a large number of images of pre-filled syringe.
Ironically, this quality, which makes a high-performance system possible, is also a defect for the biomedical industry.
"If a particle is not detected and this causes a health problem, we want to be able to trace the origin of the error to correct it. The challenge with artificial intelligence is that it trained itself. We don’t necessarily know its decision criteria.”
So the team had to invent a module to understand the system's choices and correct them if necessary. Today, the whole thing works at a very high efficiency rate.
Is Haxio about to sell its machine to other factories?
“In fact, we just made a pivot.We noticed that the mechanical aspect demanded a lot of logistics and that this could slow our growth. So today, we prefer to sell the artificial intelligence and programming that we designed and install them on other manufacturers' systems.”
And, in the end, how did Alexandre manage to find his team for all this?
“This is where Prompt helped us a lot. They put us in touch with an excellent ÉTS (École de technologie supérieure) professor who, in turn, introduced us to two students who participated in the project. One of them now works full time for us.”
AI FOR CONTAMINANT DETECTION