A human activity recognition (HAR) system carried by masseurs for controlling a therapy table via different movements of legs or hip was studied. This work starts with a survey on HAR systems using the sensor position "trouser pockets". Afterwards, in the experiments, the impacts of different parameters and configurations on the classification accuracy is examined to get a thorough understanding of the classification process.

Our research introduces an innovative method to improve the quality control of plastic granules, crucial in industries like manufacturing and automotive. We addressed a common issue: identifying "burn marks" on granules caused by overheating during processing. By combining advanced machine learning and a novel data augmentation technique called "cutout", we significantly enhanced the detection accuracy of these defects.

Tracking individual parts in manufacturing processes helps to lower production cost by assigning a full record of production steps to each part, and therefore enabling targeted quality control. A novel approach is presented, which shows that the printing quality can be monitored using images of the printed result only. Analysing data from long-term tests shows which characteristics of the part marking are suitable to be used as indicators for maintenance needs.

This paper discusses the deployment of an internet of things (IoT)-based model to monitor the fermentation of tea in a tea factory in Kenya. The model uses deep learning to detect the optimum fermentation of tea as fermentation progresses. To further improve on the results, a majority voting technique based on regions is used. The model shows promising results, and its predictions correlated well with those of the experts.

Detecting faults in bearings is indispensable for the maintenance of many industrial machines or components. The faults are detected by recognizing patterns in vibration data that are measured at the bearing housing. The method can be trained with data of variable revolution speeds, therefore reducing the effort for the acquisition of training data. Moreover, incipient faults can be detected before they cause severe damage to the equipment.