Building trust in AI
Read time: 3 minutes
Artificial intelligence can make manufacturing plants more efficient and consequently more sustainable, but it needs to be explained and humans must be able to trust in its recommendations
You have to create trust between the person and the AI.
Artificial intelligence (AI) is today’s most talked-about new technology. It has broad applications in manufacturing, from supply-chain management to quality control. Through complex problem-solving and multi-tasking, AI also improves efficiency, saving time, energy and resources and ensures more sustainable processes. However, one challenge is that it can be difficult to understand how the AI reaches its conclusions. Without that, it can be impossible to replicate a good decision, or avoid a bad one.
That’s why CNH has spent the past four years exploring the potential of explainable AI (XAI) with the XMANAI (eXplainable MANufacturing AI) project. Carried out at CNH’s San Matteo site near its plant in Modena (Italy), it enables operators to benefit from the technology while minimizing risk.
During that time, researchers made progress in using AI for predictive maintenance but, perhaps more importantly, they learnt about how AI and humans work best together. “You have to create trust between the person and the AI,” says Claudia Campanella, Head of Ergonomics & HMI (Human Machine Interface) at CNH. “That’s what brings real value and allows you to improve day by day.”
Instead of having to check lots of data to find the fault, the AI presents the probable solution as a clear visualization in a web application.
Using AI to reduce disruption
San Matteo houses one of CNH’s 49 research and development (R&D) centers, and is a hub of excellence for the engineering and technological development of our tractors globally.
Near the San Matteo R&D center, the plant in Modena produces components for small and medium tractors, including the driveline, axles and gears. Crucial to the process is the Heller MCH 400 automated machining center, which creates part of the driveline. The machine runs 24 hours a day, seven days a week, including on an empty cleaning cycle. If it breaks down, the entire production line is disrupted while operators try to find the fault. This made the San Matteo R&D site and the Modena plant the ideal location for a project exploring the use of AI for predictive maintenance and machine fault diagnosis to reduce downtime and facilitate troubleshooting.
The XMANAI project began in November 2020, with funding from the European Union’s research and innovation program, Horizon 2020, and support from the Politecnico di Milano, the University of Modena and Reggio Emilia and other companies including Deep Blue, TXT, Fraunhofer, SUITE5 and AiDEAS. Leading AI researchers at those universities and companies helped identify opportunities to deploy the latest technology.
Engineers know the machine well, but some operators and new recruits will need guidance — the AI is there to assist them, not replace them.
At the beginning, the operators were a little bit skeptical, but once they really tested the application, they appreciated its value and began to trust it.
Monitoring machine performance
The initial phase of the AI project took about a year and involved installing sensors and collecting sufficient volumes of data. Researchers used 76 different sensors to track measures such as vibration and temperature.
This data was fed to an AI model that analyzed it to find relationships between variables. The results were used for two things: first, anomaly detection, to build a picture of normal operation and provide troubleshooting assistance in the event of a fault; and second, predictive maintenance, using the understanding of machine performance to predict when a part might need replacing.
The AI uses a method known as SHAP (Shapley Additive Explanations) to explain the output of the model. This is done by both analyzing the effect of each variable on the model one at a time (a univariate basis), and by examining the combined effect of variables (a multivariate basis). The result is that the model can explain its decisions with simple graphs showing the analysis.
Giving the right answers
Engineers know the machine well, but some operators and new recruits will need guidance — the AI is there to assist them, not replace them. Instead of having to check lots of data to find the fault, the AI presents the probable solution as a clear visualization in a web application. Maintenance operators get just the basic information needed to fix the machine, while engineers, who have more expertise, get more detail — a correlation matrix showing how the anomalies match up with the data from sensors.
CNH developed an augmented reality (AR) training program for operators so that, wearing AR goggles or using a tablet computer, they could see the AI’s findings overlaid on their view of the machine. “At the beginning, the operators were a little bit skeptical,” says Campanella. “But once they really tested the application, they appreciated its value and began to trust it.”
After each use, the app asked the operator for feedback on its usefulness, intuitiveness, ease of use and their confidence in its findings. The XMANAI project concluded in April 2024 and the results were overwhelmingly positive, with 85 percent rating the app 4 or 5 out of 5 for usefulness and 67 percent giving 4 or 5 for ease of use.
The team is collating its data for publication, but it is already clear that the project has reduced machine downtime by at least 35 percent. When the team included the XMANAI web application’s suggestions which avoided the need for external maintenance calls and the subsequent wait for replacement parts, plus additional help from the AR-guided application for inexperienced operators, downtime was reduced by 75 percent. CNH is now examining ways to increase those benefits by expanding the project to other machines.
CNH is a world-class equipment, technology and services company that sustainably advances the noble work of agriculture and construction workers.
CNH Industrial N.V.
Corporate Office: Cranes Farm Road, Basildon, Essex, SS14 3AD
United Kingdom