This blog, Part 3 of a series inspired by an Advantech webinar I attended entitled Enabling Smart Factories – The Transformation of Manufacturing Systems & Factory Facilities. This post covers the second of Dr. Jay Lee’s topics, Driving Forces and Emerging Technologies (Big Data, Agent, Industry 4.0 and Cyber-Physical Systems). You can read Part 1 here and Part 2 here.
Changing Issues in Manufacturing
When it comes to the myriad of data available, manufacturing-system architects, managers and operators must decide from which machines and components they should collect how much data, under what parameters, as well as when and under what conditions. In short, data collection must be weighted, prioritized and filtered. Some components generate more and/or higher-impact issues.
Dr. Lee advises that system architects take a “worry” approach, emphasizing greater weight to the features of machines and components that have the potential of generating more or greater worries. For example, delay in the cycle time of a valve or pump might be a significant feature within a given manufacturing line necessitating greater weight to that data set.
Toyota’s Automation Vision
When Dr. Lee first worked with Toyota in 2004, its leadership foresaw a need to better integrate the actions of machines and of people. Among the challenges was that people could articulate their conditions – for example, telling managers when they felt tired – while machines couldn’t. Toyota wanted a method of assessing how and when machines degrade. The most straightforward method would be to enable machines to “feel” their degradations and to relay that information directly to managers and operators.
Dr. Lee suggests accomplishing this by modeling evaluative systems using a Predict-and-Prevent method that considers dynamic usage states rather than the more-static analysis tools of the standard Six-Sigma approach. Predictive analytics and instrumentation evaluate a specific performance feature of a component, continuously comparing the normal behavior of that feature to its most recent behavior, measuring and evaluating changes in the rates pf degradation of that feature, and extrapolating a prediction of future performance life. If properly designed and implemented, these predictive analytics will allow the component to answer the essentially human question, “How do you feel?”
Prognistics Toolbox for Manufacturing
In the future, every product will have a health indicator that transmits data continuously via a cyber-data connection. Optimization of machine performance will depend not so much on the existence of the data connection, but upon how the resulting data is analyzed – upon assignment of an appropriate algorithm to analyze the machine’s health. Dr. Lee lists algorithms commonly assigned to four toolbox functions, as follows:
- Signal Processing and Feature Extraction: Time Domain, Frequency Domain, Time-frequency and Principle Component analysis; and Fisher Criterion.
- Health Assessment: Logistic Regression, Statistical Pattern Recognition, Neural network, Feature Map Pattern Matching and Gaussian Mixture Model.
- Health Diagnostics: Support Vector Machine, Hidden Markov Model, Bayesian Belief Network and Feature Map Pattern Matching.
- Performance Prediction: Autoregressive Moving Average (ARMA), Match Matrix, Fuzzy Logic and Elman Neural Network.
Cyber-Physical Systems
Dr. Lee defines physical systems as “man-made systems governed by the laws of physics and operating in continuous time.” By comparison, cyber systems relate to discrete windows of time, involving elements of computation, communication or control that are logical and switched. When you integrate the two systems, you achieve an integrated view of how components are working within all scales and levels.
In the future, each machine may have a connection feeding data to algorithms residing in cyberspace that analyze all critical machine behaviors, building results in the form of virtual syntheses displayed as degradation models and/or risk-analysis charts, or categorized as time-machine functions. The resulting cyber system will provide an accurate representation of the machine’s physical health.
Industry 4.0
Dr. Lee identifies these cyber-physical systems as critical to enabling the fourth industrial revolution, better known as Industry 4.0. Four factors are key:
- Products will share embedded information with machines they encounter along the production line.
- Automation will assist managers and operators via machine self-optimization, self-configuration and self-diagnostic methods.
- Software applications will manage critical functions such as data acquisition, sequential and continuous control, plus trend-analysis, planning and optimization functions.
- Standards will be needed to enable inter-company networking and systems integration.
In today’s factory, smart sensors and fault-detection technologies precisely measure component status, detecting failures as they occur. In tomorrow’s factory, components will be self-aware and self-predictive, monitoring degradation for prediction of remaining useful life, anticipating problems before they occur.
In today’s factory, condition-based monitoring and diagnostics measure the quality and quality of machine throughput. In tomorrow’s factory, machines will be self-aware and self-comparative, self-predicting their respective up-times and overall health.
In today’s factory, production systems stress lean operations – optimizing output and reducing waste. In tomorrow’s factory, the goal of production systems will be worry-free productivity.