No abstract

No abstract

The still prevalent use of paper conformity lists in the automotive industry has a serious negative impact on the performance of quality control inspectors. We propose instead a hybrid quality inspection system, where we combine automated detection with human feedback, to increase worker performance by reducing mental and physical fatigue, and the adaptability and responsiveness of the assembly line to change. The system integrates the hierarchical automatic detection of the non-conforming vehicle parts and information visualization on a wearable device to present the results to the factory worker and obtain human confirmation. Besides designing a novel 3D vehicle generator to create a digital representation of the non conformity list and to collect automatically annotated training data, we apply and aggregate in a novel way state-of-the-art domain adaptation and pseudo labeling methods to our real application scenario, in order to bridge the gap between the labeled data generated by the vehicle generator and the real unlabeled data collected on the factory floor. This methodology allows us to obtain, without any manual annotation of the real dataset, an example-based F1 score of 0.565 in an unconstrained scenario and 0.601 in a fixed camera setup (improvements of 11 and 14.6 percentage points, respectively, over a baseline trained with purely simulated data). Feedback obtained from factory workers highlighted the usefulness of the proposed solution, and showed that a truly hybrid assembly line, where machine and human work in symbiosis, increases both efficiency and accuracy in automotive quality control.

In the competitive automotive market, where extremely high-quality standards must be ensured independently of the growing product and manufacturing complexity brought by customization, reliable and precise detection of any non-conformities before the vehicle leaves the assembly line is paramount. In this paper we propose a wearable solution to aid quality control workers in the detection, visualization and relay of any non-conformities, while also reducing known performance issues such as skill gaps and fatigue, and improving training methods. We also explore how the reliability, precision and validity tests of the visualization module of our framework were performed, guaranteeing a 0% chance occurrence of undesired non-conformities in the following usability tests and training simulator.

In the study of effectiveness and efficiency of an athlete’s performance, intelligent systems can be applied on qualitative approaches and their performance metrics provide useful information on not just the quality of the data, but also reveal issues about the observational criteria and data collection context itself. 2000 executions of two similar exercises, with different levels of complexity, were collected through a single inertial sensor applied on the fencer’s weapon hand. After the signals were split into their key segments through Dynamic Time Warping, the extracted features and respective qualitative evaluations were fed into a Neural Network to learn the patterns that distinguish a good from a bad execution. The performance analysis of the resulting models returned a prediction accuracy of 76.6% and 72.7% for each exercise, but other metrics pointed to the data suffering from high bias. This points towards an imbalance in the qualitative criteria representation of the bad executions, which can be explained by: i) reduced number of samples; ii) ambiguity in the definition of the observation criteria; iii) a single sensor being unable to fully capture the context without taking the actions of the other key body segments into account.