The Woodruff School

SUMMARY

Predictive maintenance is a crucial application of the Internet of Things and the digitalization of manufacturing. By predictive maintenance, manufacturers could save a total of $240 to $630 billions in 2025. One main part about the accuracy of a predictive maintenance system is the information about how fast the value of a specific manufacturing system variable changes. These changes can be described mathematically as the derivation of the examined variable. In real life, a combination of filters have to be applied to successfully evaluate the slope of the output of a variable. For this case, two major problems can be identified:
1. Noise (measurement noise and system noise) is always present in a periodic or non-periodic pattern.
2. Macrotrends are dependent on the period of examination of the test run.

To solve these problems, a set of different slope estimation algorithms are developed and tested for different use cases. The slope estimation module is able to handle the superimposed noise on a broad variety of signals as well as to detect macrotrends in data series. It can deal with different (manufacturing based) variables and handle diverse timesteps, frequency of noise, timescale of macrotrends and so forth.

The developed slope estimation module is applied on two manufacturing variables. The first one is a height measuring device for coolant in tanks. By measuring the slope of the height of the coolant, a potential increased wastage of coolant can be detected and a possibility to detect a problem is given. The second application is a pH-measurement device for coolant. By measuring the pH value of coolant, one can state the quality of the coolant. To develop a predictive method for coolant lifetime, the slope of the pH value should be determined. This enables machine users to predict the next date of coolant replacement on specific machines.

SUBJECT:	M.S. Thesis Presentation

BY:	Roman Burkart

TIME:	Friday, July 20, 2018, 7:00 a.m.

PLACE:	Love Building, 109

TITLE:	Adaptable Slope Estimation Module and its Application in a Coolant Monitoring System for Predictive Observation

COMMITTEE:	Dr. Thomas Kurfess, Chair (ME) Dr. Christopher Saldana (ME) Dr. Oliver Sawodny (University of Stuttgart) Dr. Cristina Tarin (University of Stuttgart)

SUMMARY Predictive maintenance is a crucial application of the Internet of Things and the digitalization of manufacturing. By predictive maintenance, manufacturers could save a total of $240 to $630 billions in 2025. One main part about the accuracy of a predictive maintenance system is the information about how fast the value of a specific manufacturing system variable changes. These changes can be described mathematically as the derivation of the examined variable. In real life, a combination of filters have to be applied to successfully evaluate the slope of the output of a variable. For this case, two major problems can be identified: 1. Noise (measurement noise and system noise) is always present in a periodic or non-periodic pattern. 2. Macrotrends are dependent on the period of examination of the test run. To solve these problems, a set of different slope estimation algorithms are developed and tested for different use cases. The slope estimation module is able to handle the superimposed noise on a broad variety of signals as well as to detect macrotrends in data series. It can deal with different (manufacturing based) variables and handle diverse timesteps, frequency of noise, timescale of macrotrends and so forth. The developed slope estimation module is applied on two manufacturing variables. The first one is a height measuring device for coolant in tanks. By measuring the slope of the height of the coolant, a potential increased wastage of coolant can be detected and a possibility to detect a problem is given. The second application is a pH-measurement device for coolant. By measuring the pH value of coolant, one can state the quality of the coolant. To develop a predictive method for coolant lifetime, the slope of the pH value should be determined. This enables machine users to predict the next date of coolant replacement on specific machines.