Anomaly Classification and Recovery in Industrial Machinery and Academic Applications: A Systematic Mapping based on Artificial Intelligence and Machine Learning Approaches

https://doi.org/10.5281/zenodo.20028851

Abstract

The problem of unscheduled shutdowns in the industry and Academic sectors has been of great concern and has led to serious losses in the economy, as well as the security of various sectors being compromised. Recovery of anomalies is now a major research topic of contemporary data-driven systems, even though anomaly detection is a well-researched topic, recovery mechanisms are relatively under-researched. This paper is a review paper on statistical, machine learning, deep learning, and hybrid techniques of anomaly recovery in both academic and industry environments. The paper compares the existing approaches, pinpoints the gaps in the studies and describes the trends in the future. The findings also indicate the importance of integrated detection and recovery systems in improving the performance of the systems and the reliability of the data. The increasing use of data-oriented technologies within the educational and industrial context has resulted in the development of vast amounts of data on digital platforms, sensors, and automated surveillance systems. However, these datasets are often affected by anomalies caused by sensor malfunctions, communication problems, system malfunctions, or human mistakes. These anomalies may have a major influence on the reliability of data and lower the quality of analytical models applied to make decisions. Despite detailed studies on the methods of anomaly detection, relatively less emphasis has been placed on anomaly recovery, which aims at re-creating corrupted or missing data values. This paper entails a detailed examination of anomaly detection and anomaly recovery methods based on machine learning methods. The paper discusses the classical statistical techniques for  Anomaly Classification and Recovery in Industrial Machinery and Academic Applications and present a systematic mapping based on Artificial Intelligence, Machine Learning algorithms and deep learning models that can be applied to find abnormal patterns in data sets used for Industrial and Academic Applications. In addition, we discuss reconstruction-based recovery models, such as Autoencoders, Variational Autoencoders, and Generative Adversarial Networks to fix corrupted data. The article recognizes the gaps in the existing literature on integrated frameworks of anomaly management and the necessity to integrate detection and recovery processes in one framework. The results show that machine learning models have a significant positive effect on the accuracy of anomaly detection, whereas deep learning and generative models provide opportunities to use them to recover anomalies. A combination of these may be applied to enhance the reliability of data and the functionality of the system, both in the academic and industrial contexts. Finally, the paper offers future research directions on how to develop scalable and intelligent anomaly management systems that would operate in large data environments.