Continuous asset optimization in manufacturing
Unplanned downtime is still one of the predominant costs in a manufacturing operation notwithstanding the advancements in predictive and condition-based maintenance. The focus of proactive maintenance has been to:
- Catch the symptoms of failure early
- Infer the probable causes from the symptoms using past experience and domain knowledge and,
- Take actions to prevent the failure.
However, only limited success has been achieved from the previous generation predictive maintenance approaches. Why?
This is mainly because treatment of the asset in question is a “black box.” Without the knowledge of the “internal state” of the asset and the “state transitions” that are taking place when the asset is in operation, it is not possible to infer the real cause and potential for failure. Therefore, in the past, in many instances, even if a failure was avoided in the shorter term because a symptom was treated via predictive maintenance, the asset was not “cured” from the actual cause.
A physical asset will fail unexpectedly when some failure inducing sub-optimality in its past operational behavior has gone undetected. Having said, detection of sub-optimality is not only required for avoiding unexpected failures. It is very much required for achieving superior performance of the assets.
Detecting sub-optimality and/or continuously maintaining optimal performance of assets was, at best, of theoretical interest until now. This is because the dynamics of a manufacturing shop floor made the goal of achieving continuous optimality of an asset a moving target and impossible to achieve. However, with the advent of smart manufacturing technologies, the situation has significantly changed. It is now possible to strive towards achieving “practical optimality” at an affordable cost given the current capability to generate meaningful and contextual real time information on the “internal states” of the assets.
By marrying smart manufacturing technologies with cutting edge statistical, data-driven and physics of failure modeling techniques, aiming for a new level of asset optimization is no longer impossible. This methodology is termed as continuous asset optimization (CAO).
An example of CAO using a digital twin of a foundry is illustrated in the figure below. The foundry line in this case is considered as a single asset and the optimization is applied to the entire line.
Four distinct layers of information processing are shown for achieving CAO.
The lowest layer, the operations layer, acquires operational data of the foundry. The information from the operations layer is used to analyze the performance of the asset in the performance layer.
The digital twin for CAO is vertically integrated. The operational data available in the operations layer is used for the continuous computation of the “current values” of the internal state variables. The state variables are the set of variables that define or determine the optimal performance of an asset or a sub-processes of the asset at any given time. The computed “current values” of the state variables are provided as inputs to the optimization layer.
The optimization layer is comprised of a library of models for the individual sub-processes defining the operation of the asset (foundry processes in this example). These models will present as outcome the current state of optimality of each of the sub-processes based on some objective function. From such an outcome, and the knowledge of the sequence of discrete states that are possible from the current state, the future behavior of the sub-processes are predicted using appropriate algorithms.
The combined behavior of the overall system is synthesized from the individual outcomes of the sub-processes. This forms the final prediction layer. All of the above-mentioned layers are implemented as a software platform. The approach outlined above can be applied to other types of assets such as compressors, transformers, heat exchangers, boilers and more.
Dr. Ananth Seshan is the chairman of the continuous asset optimization Working Group at MESA. This article originally appeared on MESA International’s blog. MESA International is a CFE Media content partner. Edited by Chris Vavra, production editor, Control Engineering, CFE Media, firstname.lastname@example.org.