As the manufacturing technology has been advancing and as the competition for machinery in the market mounts, the need for effective and efficient machine maintenance practices has been on the rise. Over the past few decades, efforts have been concentrated on monitoring the operational health of the machine to ensure that the performance of the machines remains at an optimal level. Machine health monitoring as a process entails observation of the condition of the machine using sensors either offline or on a continuous base. A major challenge in machine health monitoring is to eliminate redundant information and integrate multiple features to improve the efficiency and effectiveness of the process.
Recently, researchers at Xi’an Jiaotong University: Dr. Chuang Sun, Professor Ruqiang Yan, and Professor Xuefeng Chen together with Case Western Reserve University scientists Dr. Peng Wang and Professor Robert X. Gao illustrated how the performance of the manifold can be improved by eliminating the nonlinearity of the manifold. In this regard, they introduced Kernel Sparse Representation (KSR) into the manifold in a bid to better the machine. To solve the number of neighbors and connecting weights into Locally Linear Embedding (LLE), the authors used the KSR to leverage enhanced manifold learning. The research work is now published in journal, Mechanical Systems and Signal Processing. The authors identified that there is a problem with some of the current methods of manifold learning. One of the problems is that all the current methods of manifold learning lack the desired adaptability which can handle non-linear systems when variations occur. They therefore sought to counter the problem of adaptability when variations occur by developing an enhanced manifold learning method that would utilize the KSR when determining the neighbors and the weights.
In their research method, KSR for optimization involved utilization of sparse representation. This mode of representation modelled the neighborhood of the data while at the same time adaptively connecting the weights. Kernel space embedding as a process entailed several steps among them being an introduction to Locally Linear Embedding and embedding of the kernel space linearly. To monitor the health of the machine, the authors used features extracted from the frequency and time domains that are indicative of changing patterns in the vibration signal measured, as the machine transitioned from one state to another.
The authors arrived at several exciting findings. First, they identified Kernel Sparse- Locally Linear Embedding to be one of the most effective tools that can be used to monitor the health of the gearbox. However, this is not the case for all the machines considering that the latter vary from one to another. Other than determining the effectiveness of the machine, the KSR method presents itself as one of the most effective methods relative to the other methods. Although the KSR method is used to measure the health index of the machine, some limitations exist. The commonly used dimension space for presenting the data is 2-dimensional, while in reality, a 3-dimensional space would be better and more effective.
In summary, the study managed to illustrate how the identification of neighbors can be important when monitoring the health of the machine through manifold learning. Further, they showed that kernel sparse locally learning embedding can characterize both the adaptability and the nonlinearity and therefore justifies why manifold learning should be included in developing the next generation of machine health monitoring technologies.
Chuang Sun, Peng Wang, Ruqiang Yan, Robert X. Gao, and Xuefeng Chen. Machine Health Monitoring Based on Locally Linear Embedding with Kernel Sparse Representation for Neighborhood Optimization. Mechanical Systems and Signal Processing. 114 (2019) 25–34Go To Mechanical Systems and Signal Processing