Research on a Defecation Pre-Warning Algorithm for the Disabled Elderly Based on a Semi-Supervised Generative Adversarial Network.

The elderly population in China is continuously increasing, and the disabled account for a large proportion of the elderly population. An effective solution is urgently needed for incontinence among disabled elderly people. Compared with disposable adult diapers, artificial sphincter implantation and medication for incontinence, the defecation pre-warning method is more flexible and convenient. However, due to the complex human physiology and individual differences, its development is limited. Based on the aging trend of the population and clinical needs, this paper proposes a bowel sound acquisition system and a defecation pre-warning method and system based on a semi-supervised generative adversarial network. A network model was established to predict defecation using bowel sounds. The experimental results show that the proposed method can effectively classify bowel sounds with or without defecation tendency, and the accuracy reached 94.4%.

A Prediction Model of Defecation Based on BP Neural Network and Bowel Sound Signal Features.

(1) Background: Incontinence and its complications pose great difficulties in the care of the disabled. Currently, invasive incontinence monitoring methods are too invasive, expensive, and bulky to be widely used. Compared with previous methods, bowel sound monitoring is the most commonly used non-invasive monitoring method for intestinal diseases and may even provide clinical support for doctors. (2) Methods: This paper proposes a method based on the features of bowel sound signals, which uses a BP classification neural network to predict bowel defecation and realizes a non-invasive collection of physiological signals. Firstly, according to the physiological function of human defecation, bowel sound signals were selected for monitoring and analysis before defecation, and a portable non-invasive bowel sound collection system was built. Then, the detector algorithm based on iterative kurtosis and the signal processing method based on Kalman filter was used to process the signal to remove the aliasing noise in the bowel sound signal, and feature extraction was carried out in the time domain, frequency domain, and time-frequency domain. Finally, BP neural network was selected to build a classification training method for the features of bowel sound signals. (3) Results: Experimental results based on real data sets show that the proposed method can converge to a stable state and achieve a prediction accuracy of 88.71% in 232 records, which is better than other classification methods. (4) Conclusions: The result indicates that the proposed method could provide a high-precision defecation prediction result for patients with fecal incontinence, so as to prepare for defecation in advance.

Robust seam tracking via a deep learning framework combining tracking and detection.

To address the problem of low welding precision caused by possible disturbances, e.g., strong arc lights, welding splashes, and thermally induced deformations, in complex unstructured welding environments, a method based on a deep learning framework that combines visual tracking and object detection is proposed. First, a welding image patch is directly fed into a convolutional long short-term memory network, which preserves the target's spatial structure and is efficient in terms of memory use, with the aim of avoiding some disturbances. Second, we take advantage of features from various convolutional neural network layers and determine weld feature points through similarity matching among multiple feature layers. However, feeding in noisy images causes the tracker to accumulate interference information, which results in model drift. Thus, using a welding seam detection network, the object filter is periodically reinitialized to improve tracking accuracy and robustness. Experimental results show that the welding torch runs smoothly with a strong arc light and welding splash interference and that tracking error can reach ±0.5mm, which is sufficient to satisfy actual welding requirements. The advantages of our algorithm are validated through several comparative experiments.

An Adaptive Sliding-Mode Iterative Constant-force Control Method for Robotic Belt Grinding Based on a One-Dimensional Force Sensor.

To improve the processing quality and efficiency of robotic belt grinding, an adaptive sliding-mode iterative constant-force control method for a 6-DOF robotic belt grinding platform is proposed based on a one-dimension force sensor. In the investigation, first, the relationship between the normal and the tangential forces of the grinding contact force is revealed, and a simplified grinding force mapping relationship is presented for the application to one-dimension force sensors. Next, the relationship between the deformation and the grinding depth during the grinding is discussed, and a deformation-based dynamic model describing robotic belt grinding is established. Then, aiming at an application scene of robot belt grinding, an adaptive iterative learning method is put forward, which is combined with sliding mode control to overcome the uncertainty of the grinding force and improve the stability of the control system. Finally, some experiments were carried out and the results show that, after ten times iterations, the grinding force fluctuation becomes less than 2N, the mean value, standard deviation and variance of the grinding force error's absolute value all significantly decrease, and that the surface quality of the machined parts significantly improves. All these demonstrate that the proposed force control method is effective and that the proposed algorithm is fast in convergence and strong in adaptability.

Learning Siamese networks for laser vision seam tracking.

To design a stable laser vision seam-tracking system, an advanced weld image processing algorithm based on Siamese networks is investigated and proposed to resist the interference of arc and spatter in the welding process. This specially designed neural network, combined with powerful feature expression capabilities of deep learning, takes two welding images with different sizes as inputs and generates a target confidence map in a single forward pass by using the cross-correlation algorithm. To prevent the error accumulation and model drift, an online update strategy via local cosine similarity is developed. The use of metal inert-gas welding can realize real-time and precious tracking under the condition that the strong arc continuously shields the welding seam feature points.

A study on robot force control based on the GMM/GMR algorithm fusing different compensation strategies.

To address traditional impedance control methods' difficulty with obtaining stable forces during robot-skin contact, a force control based on the Gaussian mixture model/Gaussian mixture regression (GMM/GMR) algorithm fusing different compensation strategies is proposed. The contact relationship between a robot end effector and human skin is established through an impedance control model. To allow the robot to adapt to flexible skin environments, reinforcement learning algorithms and a strategy based on the skin mechanics model compensate for the impedance control strategy. Two different environment dynamics models for reinforcement learning that can be trained offline are proposed to quickly obtain reinforcement learning strategies. Three different compensation strategies are fused based on the GMM/GMR algorithm, exploiting the online calculation of physical models and offline strategies of reinforcement learning, which can improve the robustness and versatility of the algorithm when adapting to different skin environments. The experimental results show that the contact force obtained by the robot force control based on the GMM/GMR algorithm fusing different compensation strategies is relatively stable. It has better versatility than impedance control, and the force error is within ~±0.2 N.