[Factory Control | Factory Analytics | Factory Design | Miscellaneous]
Factory Analytics

   - In-Process Monitoring and Control

  • Artificial Intelligence for In-Process Quality Inspection and Process Control
  • Failure Diagnosis & Prognosis for Machine Healthcare

Feature Engineering & Process Fault Detection:


Feature Engineering & Process Fault Detection

Anomaly Detection: Industrial manufacturing system typically have large amounts of data generated by sensors and other monitoring equipment. Anomaly detection and fault diagnosis are important techniques used in the industrial sector to monitor and detect abnormal conditions in complex systems such as machines, processes, and equipment. These techniques aim to identify potential problems early on and prevent equipment failure or breakdowns. Anomaly detection involves monitoring a system for abnormal behavior or patterns that deviate from expected or normal behavior. It involves collecting data from sensors and analyzing it using statistical and machine learning algorithms to detect deviations from normal behavior. For example, in a manufacturing plant, anomaly detection algorithms can be used to detect unusual fluctuations in temperature, pressure, or vibration that may indicate a potential problem.


Anomaly Detection

Anomaly Detection by Deep Learning: There are large amounts of data generated by sensors and other monitoring equipment in industrial manufacturing systems. Anomaly detection is performed by applying various techniques to the data, and in particular, anomaly detection models using deep learning have been mainly used recently. Deep learning can be used to learn complex hierarchical feature relationships within high-dimensional raw input data to achieve better performance in anomaly detection systems.[3] Also, depending on the characteristics of applied data, the type of model used is divided into semi-supervised, unsupervised, hybrid, and one-class neural networks.


Anomaly Detection by Deep Learning

Failure diagnostics and redesign protocol:                                                                                                                                                                                                                                       

Failure diagnostics

Design from failure: Despite several pre-validation efforts, unknown faults and quality degradation occurs all too frequently, often because of the challenges designers and engineers face in trying to predict every customer's behavior of product usage and the outcomes of sub-components interaction in a product or system. In addition, there is no sufficient discussion on redesign procedures using failure analysis results. Therefore, the overall objective of this research is to develop an architecture beyond just fixing failures, in order to not only analyze the root causes of NFF failure and diagnose and prevent the failure but also connect the relevant information and knowledge including analysis results to redesign products or systems efficiently. More...

Design from failure

Event-driven machine failure analysis: Today's data acquisition and storage technologies have enabled manufacturers to acquire product field performance as well as manufacturing process information in the form of event logs containing massive datasets. Events are recorded in response to system state or operation changes such as critical value changes in parametric data. Failures can be defined as system faults or performance degradation, which are strongly related with events or directly parametric data. Events are triggered by value changes in the corresponding sensor data. Therefore, optimal strategies for the diagnosis of product field failures will be developed. The advantages of the event-driven failure analysis approach are (i) cost effective for pattern extraction and matching as compared to parametric data-driven approaches, (ii) capability of including functional interactions of complex systems and user behaviour for failure analysis and (iii) systematic and hierarchical reasoning of the root causes of failures. More...

Design from failure

Sound source localization based quality monitoring: The importance of product quality control cannot be overestimated, and effective quality control starts from the definition of appropriate key performance indicators for a product. In many cases, acoustic noise creates a critical quality problem, for example, annoying squeak and rattle sound from a car door can cause a significant warranty failure. In this research, we are developing a noise detection and localization method in a clamorous manufacturing environment. Sound source localization and noise cancelling methods are employed to identify the precise location of noise source. More...

Sound source localization

©2020 Factory Intelligece Lab POSTECH
Cheongam-Ro 77, Pohang, South Korea, 37673
Tel:+82 54 279 8261 Fax:+82 54 279 2870