LET’S TALK: IIoT Enabled Testbeds & Big Data Analytics Applications for Smart-Manufacturing
Abstract:
Candidate summary:
· Creative engineer with 2 years of industrial experience and 4+ years of research experience in IIoT applications, big data analytics, smart manufacturing, process modelling & optimization resulting in 4 successful collaborative research projects & 7+ peer-reviewed publications.
· Strong algorithm and programming experience in MATLAB, python, Linux high performance computing environments and parallel computing evidenced by numerous written public and private codes during execution of research projects. Public codes are available on GitHub (link on poster).
· Recognized leader & Effective communicator demonstrated by 10+ presentations at international / national conferences and through dedicated service in diverse organizations.
Poster summary:
Why IIoT & Big data analytics?
They enables real-time (remote) monitoring and control, predictive maintenance, etc. These capabilities help improve plant efficiencies by achieving better control, reducing off spec production, reducing equipment failures and process down time, etc.
What are our contributions?
We’ve made contributions to all aspects of smart manufacturing. From hardware to software, from data collection, data management, to data analytics. We have demonstrated how to make standard unit operations smart by utilizing noninvasive IIoT sensors, Raspberry Pi microcontrollers, lightweight wireless communication protocols, systems engineering enhanced machine learning algorithms, model based feature selection, data mining etc.
Specific contributions:
1. IIoT enabled industrial testbeds:
Every step involved in developing IIoT enabled testbeds will be discussed. We have tested IIoT temperature sensor, vibration/accelerometers sensor (both digital and analog), near IR light sensor, IR camera, video camera etc.
Key discussion points:
- Lab setup, experimental design & IIoT data characteristics & challenges.
- Predictive modelling framework built with brief description of data compilation, data mining, and data analysis procedures.
- Novel model based data filtering or feature selection approach capable of selecting relevant information for vibration datasets.
- Performance comparison of hierarchical linear modelling with neural networks (ANN & LSTM).
2. Statistics pattern analysis (SPA) modelling for spectrum data:
We will discuss SPA based modelling approach developed for spectrum data. Developed approach captures more information with less computation ideal for smart-manufacturing applications.
Basic Idea:
- Divide spectrum in different intervals and calculate statistics for each.
- Identify most relevant statistics and use as independent variables.
Scope:
- Tested on 4 public real world NIR, UV/Vis spectrum data from different industries.
- Compared with 4 other linear, non-linear regression approaches
Major Advantages:
- Better prediction accuracy with improved robustness.
- Simpler model with few important variables.
- Incorporates non-linearity.
- Faster training and validation.
3. Statistics pattern analysis (SPA) modelling for batch data:
Next a variant of SPA modelling approach is presented for semiconductor batch data. Proposed variant was proven to have better performance by comparing its performance with other approaches used in semiconductor industry.
Proficient with the following data analytics and modelling/optimization techniques:
Partial least squares (PLS), Kernel-PLS, Synergy interval partial least squares, Artificial neural networks, Long short term memory (LSTM) neural networks, Binary matrix PLS, Principal component analysis, Time series analysis, Recursive PLS, Kalman filter, LASSO, Bayesian optimization, Gaussian process, Data filtering, Variable selection, Lomb’s Algorithm, Fourier analysis, Signal processing.
See more of this Group/Topical: Spring Meeting Poster Session and Networking Reception