Big-data-driven anomaly detection in industry (4.0): An approach and a case study

Abstract

In this paper we present a novel approach for data-driven Quality Management in industry processes that enables a multidimensional analysis of the anomalies that can appear and their real-time detection in the running system. The approach revolutionizes the way how quality control (and esp. anomaly detection) will be realized in production processes influenced by many parameters that can be in complex nonlinear correlations. It consists of two main steps: learning the normal behavior of the system (based on past data) and detecting an anomalous behavior in the real-time (by processing real-time data). The approach is especially suitable for modern industry systems that follow Industry 4.0 principles of ubiquity sensing and proactive responding. One of the main advantages is the self-adaptive nature of the approach due to its data-driven orientation, so that the model and parameters of the approach will be continuously updated to the dynamicity of data. The approach has been applied in the process of manufacturing microwave ovens (Whirlpool) and in this paper we present results for the data-driven quality control of one of the most critical parts — microwave oven fan. Due to the high speed of the rotation, every item has to be very precisely produced (according to the CAD model), which requires very strong quality control process

Towards Large-Scale, Heterogeneous Anomaly Detection Systems in Industrial Networks: A Survey of Current Trends

“…Physical properties of the process are constantly monitored, often polling data every few milliseconds in the case of critical variables, which with large, continuous processes can lead to a scenario where it is necessary to use Big Data Analytics (BDA), covered in Section 1, in order to process field and control data. This is further confirmed by proposals that, outside the field of security research, point to this need and propose several BDA solutions focused on industrial applications, such as process monitoring [51][52][53][54], maintenance [55], fault detection [56], and fault diagnosis [57,58].…”

Anomaly detection in Skin Model Shapes using machine learning classifiers

“…Moreover, in an online inspection, change detection of process distributions is difficult to monitor using statistical process control methods, as the number of samples is significantly smaller than the number of surface points to extract an accurate process distribution [22]. Point cloud-based change detection approaches focus computing distance between two registered point clouds, e.g., [19,23].…”

Evaluation of orientation on performances and process improvements in manufacturing organizations

“…Organizacije orijentisane na upravljanje poslovnim procesima mogu koristiti uobičajene metodologije upravljanja poslovnim procesima za određena poboljšanja; međutim, upotreba ovih metodologija ne implicira da je organizacija posvećena praksi upravljanja poslovnim procesima [10]. Jednim od najvažnjih izazova u proizvodnoj delatnosti navodi se kontinuitano poboljšanje procesa koje zahteva nova shvatanja o ponašaju procesa ili kontroli kvaliteta procesa radi razumevanja potencijala optimizacije ili poboljšanja [11].…”