Smart connected worker in advanced manufacturing

Funded by DOE/CESMII, the smart connected worker system (SCW) that utilizes vision based technology to characterize workflows that can be effectively combined with other sensor data; wireless mobile sensor system for electromagnetically noisy industrial environments; machine learning algorithms to model the correlated states of energy consumption and smart worker workflows.

A web app for workflow monitoring along with backend data streaming, storage, query, and machine learning data procesing is buit as shown on left. Paper 1 Paper 2

Energy disaggreation of industrial machine components

We developed a method to detect and classify power events and conduct unsupervised energy disaggregation of individual machine functioning components. The disaggregated component states are correlated with contextual information to be used for non-intrusive machine monitoring and context-aware anomaly detection. Paper

Adaptive Learning of worker action recognition

An adaptive machine learning (ML) based smart manufacturing interactive cyber physical human system (ICPHS) is designed. The ML model during deployment self-evolves with the streaming data in a self-labeling manner. The system defines a causal and temporal mapping of worker and machine states where one side can label the other automatically. The fully automated adaptive ML system can improve its accuracy for human machine interaction detection and shows potential for class incremental learning. Paper

Predictive maintenance of vacuum pumps using power signals.

We use low-frequency power signal to analyze the performance of vacuum systems over a long time span. Key features regarding pump performance, such as pumping down duratinon and peak power, are extracted for trend analysis and PM. ML methods are applied to predict the breakdown.

OCR for 3-D printers by deep learning models

The non-standard machine interfaces always make it difficult to directly pull data from 3-D printers. We developed non-invasive OCR for 3-D printers by using traditional CV and deep learning models (CRAFT and CRNN) to locate the control panel and recognize the texts on the screens in real time. Paper

RISC-V Based Object Tracker

A demonstration project designed and constructed by using a PIXY camera and a Microsemi SmartFusion2 FPGA (Future Electronics Creative Board) soft implementation of the RISC-V processor. This demo used a green object as the training target to track and follow. Video and Github Code.