AI on its way into footwear factories

Research institution INESCOP has shared details of some of the practical experience it has built up so far of using artificial intelligence in working shoe manufacturing facilities.

Talk about artificial intelligence (AI) may be all the rage but, beyond that, examples have now come to light of the technology in use on footwear factory floors, contributing to the shoe production process. These are early, tentative steps, but the industry’s transition towards becoming more automated and digitalised is definitely under way now.

Software and AI engineer Carlos Vélez García is part of the robotics and automation team at Elda-based footwear research institute, Inescop. He used a speaker-slot at Calzatic, a recent conference in Elche, to share his thoughts on using AI in footwear production and to flag up some wins in this area that Inescop has been able to chalk up so far.

He took his own first steps into this subject matter around six years ago when he was completing his final project for a postgraduate degree devoted to AI. As part of this, he worked with a major retail group in Spain on a project to make shopping easier for customers with a visual impairment, designing an app to help them find their way round a store.

Using AI-enabled sensors inside the store, the app was also able to help determine users’ locations so as to give them concrete instructions through a voice interface for making their way to the footwear department, the menswear department or wherever they wanted to go next. He points out that, if anyone should think this a small-scale project, it took place at a time when the world of business was even less familiar with AI than it is now. “This was in 2018,” he says. “No one had heard of ChatGPT.”

Shoe subtleties

All industries have their specific demands, but Mr Vélez’s opinion is that there are subtleties in footwear production, an industry that has traditionally been so reliant on the skill of manual workers, that make it particularly tricky to integrate artificial intelligence into its factory set-up. “My view is that you could almost talk about each shoe as a work of art,” he says, “and this is down to the skill and experience of the workers. You have to wrestle with a wide range of materials that can often seem the same at first glance. When you pick them up to work with them, though, you realise that there are often differences in the texture or in the exact colour.” With this, and with seasonal changes that still matter a great deal in footwear collections, shoe factories face constant change.

It is this constant change, he argues, that makes it difficult to apply large-scale automated systems to footwear production in a similar way to the use of those systems in, say, automotive or pharmaceuticals. He also makes the point that the footwear manufacturing landscape is made up mostly of small- and medium-sized companies that are usually less able to take on big technological transitions. But he insists that it is important to try to face up to these challenges.

The footwear sector should do this step by step and focusing on low-hanging fruit, he suggests, to help the industry keep functioning well as it confronts an even greater parallel challenge, that of recruiting new generations of shoe workers. “I know you don’t need me to tell you that,” Carlos Vélez said to the footwear manufacturers in the Calzatic audience. “But it means we are going to have to think about AI.”

Early harvest

In the work that Inescop has been able to do on this so far, low-hanging fruit has manifested itself as AI-enabled pliers. These were one of the results of a recent project called SoftManBot, which Inescop developed with a soling footwear manufacturer and a materials supplier in and around Elche, with financial support from the European Union. The specific context was the task of taking soles from moulds. “To do this, you have to use quite a lot of force, but you also need to be skilful,” the software engineer says. “I got to try it and I think it was immediately clear that I was using too much force and not enough skill. The result was soles that would break or end up out of shape.”

When he asked the experienced workers around him to explain exactly how they went about this part of the job, they were unable to. They said they could feel from the way the material responded to contact with the pliers what they needed to do next. The solution was to integrate movement-tracking sensors and LED lights into the pliers the workers use. All they had to do was remove the soles as normal, doing it exactly as they always had done, and as they did so, they programmed the robotic demoulding system. Trying to put the technology into the moulds would have been much trickier; the moulds are one of the aspects of footwear production that change with great frequency. AI applied to pliers attached to a robotic arm worked more effectively, combining software, hardware and AI together, with the AI supplying soft skills that we have been unable to achieve with technology up until now.

A helping hand

Inescop’s vision for AI is not that it should replace human workers in factories, but help them, Carlos Vélez explains. The aim with the system that emerged from the SoftManBot project is not to take people’s employment away, but to relieve them of repetitive and physically demanding aspects of their work, making their working lives less physically exhausting. When the injected soling materials cool down to the point at which the soles can be removed from the moulds and remain intact, workers have to apply plenty of force. The research institute has measured this effort as being equivalent to lifting, albeit briefly, a weight of between 20 and 30 kilos. The workers may not have to sustain that effort for long each time, but having to apply it every 40 seconds across an eight-hour shift is physically demanding.

A second project for which he has high hopes has the same objective: using AI to automate the initial stages of defect-detection systems in the factory. Its name is QRAIS. This can free up a quality-control manager from staring at images on a screen for an entire shift. Visual fatigue in that person can lead to problems going unnoticed until shoes reach shops or customers’ feet. The solution involves putting AI into the X-ray machines that many factories already use for quality control. The new system has the ability to detect a variety of potential problems, ranging from a pair of size 36 shoes having been packed into a box for size 30, to heel defects and even nails or staples presenting an injury-risk. If the system identifies any of these problems, it alerts the operative immediately so that a thorough visual check can take place. The technology has an easy starting point. The shoes in a pair should be mirror images of one another. AI can tell immediately if there is anything in one of the shoes that is not present in the other.

Physical demand

An early version of this system went into action in a local shoe factory in early 2024, with one of the benefits being that this exposure to the real world has enabled the technology to improve. It is gathering data all the time on what is and what is not a defect, making it better at flagging up genuine problems. A further important benefit of this adaptability is that, as the experienced workers in the factory either correct or affirm the system’s notifications, it is their way of working that the tool is learning and falling into line with. This means the technology is becoming increasingly personalised to suit their particular needs and demands.

Increasing the technological level required while alleviating some of the physical demand may be one way of attracting younger people into the industry and making sure recruits stay in the sector for longer. At the same time, AI can help footwear manufacturing companies digitalise the skills and knowledge that the current generation of workers has in abundance. “It is of fundamental importance to carry out that digitalisation effort and to start right away,” Carlos Vélez says. “If we don’t, the day will come when those workers retire and walk away from the industry taking that knowledge with them.”?

SoftManBot tracked the different movements and levels of force that workers use to extract footwear soles from moulds in order to help automate the task. All credits: Inescop