In the automotive sector, ensuring the integrity of every component is crucial, both to ensure vehicle safety and to optimize customer satisfaction.

However, traditionally, many quality control processes are carried out manually and can therefore be inefficient: subject to human error, variable in the precision of checks, not always sufficient to identify all possible anomalies.

One of our partners in the automotive sector, for example, had a department dedicated to manual inspections of rubber components, to certify the integrity and safety of finished products before putting them on the market.
The inspections, carried out on more than 2000 pieces every day, involved a very rapid visual check, which took place by manually moving and rotating the pieces and, therefore, trying to identify any flaws in a few milliseconds.

Overall, it was an alienating and exhausting job for the operators, but also highly imprecise, susceptible to frequent oversights and errors.
The very tight work pace barely kept up with the number of pieces produced daily but gave operators only a few moments to carry out the checks, limiting their accuracy and depth.
Furthermore, some limitations, in terms of time and physical capacity, reduced the possibility of examining the component from the inside, preventing the inspection of some areas of the product prone to rather recurrent defects, such as folds and internal membranes.

The company essentially had a very expensive, slow, and inaccurate Quality Control process: modest in detecting the most critical defects, lacking in offering guarantees about the full safety of the produced pieces.

To address the customer's problem, our team developed a solution to automatically perform visual inspections on rubber components.

First of all, given the need to implement controls inside the components as well, we collaborated with a company specialized in hardware development to install specific fiber optic systems in the control lines to capture images from inside the inspected pieces.

Then we developed an automatic control system, intertwined with the cameras (both existing ones and fiber optic ones) and connected to a Cloud platform for monitoring and analysis.

The platform allows monitoring in two phases: an initial, real-time phase, which allows observing minute by minute the inspections carried out by the AI; a subsequent analysis phase, where the reasons and causes of any production flaws can be investigated.

Defect detection occurs thanks to the integration of the platform with a “hybrid” Machine Learning model, which combines the power of Computer Vision for image analysis with the capabilities of Anomaly Detection in detecting anomalies.
The software starts by screening the images collected in real-time, extracts sections to segment and analyze, and then verifies their condition by comparing “normality” and “anomaly”. The algorithms, in fact, during the training phase, learn to recognize the “norm” of materials and shapes and, during monitoring, identify “anomalous” data, which differ from the already learned patterns and could indicate discrepancies, defects, or inaccuracies.

The sophistication of the anomaly detection thus constituted allows identifying not only recurring anomalies but also random and unpredictable defects, increasing the accuracy of inspections and minimizing false positives detected.

In this way, our customer's quality controls have become continuous, faster, and more accurate.
Over time, this solution has helped reduce overall production costs and increase the quality level of the offered products, with greater satisfaction also of the human capital, reallocated to less alienating and repetitive tasks.