Activiteit

Drawing inspiration from existing silicone-based products, this report presents and compares three sensor designs: two capacitive and one employing magnetic principles, all based on silicone. Moreover, research was conducted on a complete pipeline for the purpose of gathering and storing sensor data. The calibration machine was instrumental in conducting compression tests with forces ranging from 0 N to 300 N. Local assessment of sensor responses in different regions was also a part of the process. The three sensors, after being tested, were put through a comparative process with the outcomes from a solid silicon pad. Among the capacitive sensors, one replicated the performance of solid silicon, while the other two showed either poor repeatability or extreme rigidity, which could compromise patient comfort. In conclusion, the proposed system effectively showcased its capability to quantify and track orthosis-applied forces, thereby validating its suitability for clinical use.

Parkinson’s disease (PD) frequently presents with freezing of gait (FoG), a debilitating clinical condition marked by the inability to progress forward while intending to walk. This is one of the most bothersome signs of Parkinson’s Disease, which contributes to a heightened risk of falling and a lower quality of life. Monitoring FoG in real-world scenarios is facilitated by a workable combination of wearable inertial sensors and machine learning algorithms. While traditional Fog of War (FoG) detection algorithms examine all data without discrimination, the contextual activity is absent from the analysis. The focus of this study was the development of a lightweight, context-sensitive algorithm that only activates FoG detection systems when required, thereby mitigating computational overhead. The project saw the implementation of diverse gait recognition approaches, encompassing machine learning (ML), deep learning (DL), and a singular acceleration magnitude thresholding strategy. To evaluate and train context algorithms, inertial sensor data from three distinct datasets were extracted, encompassing a total of eighty-one Parkinson’s Disease patients. wnt signals receptor Recognition of gait, utilizing a one-dimensional convolutional neural network, showcased sensitivity ranging from 0.95 to 0.96 and specificity from 0.80 to 0.93. The performance of the threshold-based approach for evaluating context awareness in FoG detection surpassed that of machine learning and deep learning. Algorithms designed with context in mind facilitate the removal of over 55 percent of non-FoG data, retaining all but less than 4% of FoG occurrences. The efficacy of fog detection algorithms can be improved through the application of a context classifier, achieving significant computational savings without compromising accuracy in fog detection. Hence, the application of context awareness furnishes a solution for energy-efficient, long-term FoG observation in ambulatory and free-living circumstances.

The integration of convolutional and transformer networks within hybrid models yields impressive outcomes for human pose estimation. However, the current hybrid models in human pose estimation, which generally use self-attention stages following convolutional ones, are liable to internal disagreements. The conflict between modules dictates one module’s dominance in these hybrid sequential models. Consequently, the performance of precisely determining the positions of more refined keypoints does not match the overall performance. We developed a hybrid parallel network to alleviate this shared conflict, achieving this by parallelizing the self-attention and convolution modules, allowing us to make optimal use of their respective strengths. The self-attention branch, operating within the parallel network, primarily models long-range dependencies to improve semantic representation, while the convolution branch, with its local sensitivity, simultaneously enhances precise localization. We proposed a cross-branch attention module to further minimize the conflict, by regulating the features from both branches along the channel axis. A 756% and 754% AP improvement across the COCO validation and test-dev sets was achieved by the hybrid parallel network, maintaining consistent performance in high-precision localization and overall performance. The experiments show our hybrid parallel network to be on a par with the current best human pose estimation models.

Proper offloading treatment plays a vital role in promoting healing within diabetic foot ulcers. Objective monitoring is recommended for assessing adherence to offloading protocols. Although this may be the case, self-reported adherence is commonly used, with its validity and reliability remaining unknown. This study was designed to evaluate the trustworthiness and consistency of self-reported usage of removable cast walker (RCW) offloading treatment among individuals with diabetic foot ulcers (DFUs). The study cohort included 53 participants who had DFUs and employed RCWs. Self-reported adherence rates for the proportion of daily steps attributable to the RCW were collected from each participant. Adherence of participants was ascertained objectively with the help of dual activity monitors. One week later, a sample of 19 participants self-reported their percentage of adherence, which was used to evaluate the test-retest reliability of the measurements. Validity testing involved the use of Pearson’s r and Bland-Altman plots, and Cohen’s kappa was used to determine reliability. Objectively measured adherence was notably lower than self-reported adherence, with a significant difference observed (35% (19-47) versus 90% (60-100), p < 0.001) when comparing the median and interquartile range of self-reported values. The study found substantial agreement (r = 0.46; p < 0.001), alongside extensive 95% limits of agreement, indicating a noticeable proportional bias (β = 0.46, p < 0.001), signifying validity. However, test-retest reliability demonstrated only minimal agreement (κ = 0.36; p < 0.001). Regarding individuals with DFU, the validity and reliability of self-reported offloading adherence are, at most, only fair. A pronounced overestimation of offloading adherence is commonly observed among those with DFU. Objective adherence measurement tools should be employed by clinicians and researchers, replacing subjective reporting.

This research presents a novel system for the immediate warning of CNC lathe faults and the determination of the origin of those faults. A system for obtaining information was conceived, based on the analysis of the interconnected mechanical parts in CNC lathes. Following the denoising and coarse-graining of the collected status signals, the transfer entropy theory was employed to compute the net information transfer entropy between the mechanical components, subsequently leading to the construction of an information transfer model. By utilizing the sliding window methodology, the probability threshold interval of the net information transfer entropy for lathe mechanical components under various processing configurations was established. Accordingly, the information entropy-based determination of the transition critical point enabled a detailed understanding of the fault development process. Information transfer variations between parts were instrumental in achieving fault early warning and fault root tracking for the CNC lathe. The proposed method empowers fault diagnosis with digitalization and intelligentization, leading to a timely and efficient diagnostic process. Finally, the experiment using a numerical control lathe tool to process parts validates the efficiency of the proposed approach.

The relationship between cutting force in lathe work and tool wear directly correlates with the resulting turning quality. The endeavor of directly measuring the cutting force by monitoring the strain within the tool holder is complex, as the tool holder’s design deliberately seeks maximal rigidity to enable the management of significant cutting forces. For this reason, the prevailing dynamometer designs modify the standard tool holder by lowering its structural rigidity, thereby compromising the accuracy of the machining process and leading to its limited adoption. This paper successfully detects ultra-low strain in the tool holder, crucial for maintaining machining precision, using highly sensitive semiconductor strain gauges (SCSGs) close to the blade cutting insert, thereby mitigating the effects of the dynamometer’s low stiffness. Consequently, the high temperature coefficient of SCSG caused significant readout drift due to the substantial temperature increase induced by the cutting process, which heated the tool holder’s force measuring area by approximately 30°C. For resolving this problem, continuous monitoring of the tool holder’s temperature was essential, and a BP neural network was employed to compensate for the drift errors originating from temperature. Our methodologies have refined the BP neural network prediction, achieving a sensitivity of 114 10-2 mV/N and a reduced average relative error of 148%, while preserving the original stiffness of the tool holder. The recently developed smart tool holder boasts a high natural frequency (6 kHz), making it ideally suited for dynamic cutting-force measurement. In lathe cutting experiments, the gathered data show comparable results to measurements using traditional dynamometers. The resolution of the smart tool holder is a precise 2 N, accounting for 0.25% of the total range.

The intricate relationship between gait and balance arises from the interconnectedness of the brain, nervous system, sensory organs, and musculoskeletal system. The type of footwear, manner of walking, and nature of the surface profoundly impact their actions. This exploratory study investigates the impact of Infinity Walk, pronation, and footwear parameters on the effective connectivity of the brain. A continuous-wave near-infrared spectroscopy device for functional imaging captured data from five healthy individuals.