Data driven methods for non-disruptive digitalisation of real manufacturing environments
Publication Date
July 31, 2023
Creator
Abstract
As a result of current trends towards digital transformation and the integration of Industry 4.0 ideals and technologies, many manufacturers are racing to stay competitive in an increasingly digitised environment. Among them is Siemens Congleton, a manufacturer of variable speed drives and the funding body for this research project. This factory wants to maintain competitiveness in a changing manufacturing environment, but with minimal disruption to existing processes.
Initial assessments of the factory processes show it to use many modern techniques and technologies. Despite this, some areas of production still have scope to be improved. The first of these is the scheduling method in use: a Priority Rule Based Method (PRBM). An alternative has been proposed: a Mixed Integer Linear Programming (MILP) algorithm, capable of creating a schedule that accounts for the sequence dependant setup times present in this manufacturing process. This is tutorialised for non-technical readers and compared to a simplified version of the factory’s scheduling method, on a simplified factory environment. Formulations of the algorithm are presented for single, serial and parallel setups, allowing for a modular approach to constructing a global scheduling method as well as local schedules for specific machine groups. In almost all ways the optimisation approach is better than or at least equal to the PRBM. The main drawback is the increased time taken and processing power required: however, this can be overcome by performing the calculations on a cloud platform and taking advantage of elastic computing technologies.
Another aspect of the Siemens factory that was identified as an opportunity to improve through digital transformation was the reactive maintenance and fault response procedures surrounding legacy machines. These machines cannot natively connect with wireless networks and create local data silos. As such, a generic Fault Detection and Health Monitoring (FD-HM) method using indirect data collection has been developed. It is presented in the context of a wave soldering machine use case, which suffers from solder build up in the pipes. A simple test bench is used to mimic the behaviour of the machine. Using an accelerometer and a Raspberry Pi, vibration data is collected. A gate valve creates the pipe blockage and a flow meter records the effect on the flow rate. This is then analysed using Bayesian Change point Detection (BCD) to detect and isolate the pump response when pipe blockages form. The results of the analysis are compared to the flow rate to asses the algorithm. The information can be used to identify the pump response in the future and inform machine users that maintenance is required. This reduces both the time necessary for establishing the presence of the machine fault, and the time spent using the machine while it is in that state.
This is followed by an online version of the BCD (OBCD) algorithm and an assessment of its effectiveness. By collecting incoming data points and comparing them to the observed data already collected during the current cycle, the likelihood of a change point taking place can be assessed. This algorithm is intended to provide a real time indication of the machine state to operators.
Finally, the implementation and use of both the scheduling algorithm and FDHM methods in a factory environment is discussed. A comparison between the MILP algorithm and the factory PRBM is presented with anonymised production times. An initial analysis of factory data is performed using the BCD and OBCD algorithms alongside the analysis of artificial data sets representing different potential machine fault scenarios. In addition to this, integration with Siemens’ Industrial Internet of Things (IIoT) service, MindSphere, will be considered. To conclude, the application of a combined scheduling and FD-HM approach is presented and discussed.
Item Type
ethesis
Thesis Type
PhD
Subjects (LC)
Associated Schools / Departments
Department of Mechanical, Materials and Manufacturing Engineering (UK)
eprints ID
74000
UoN Repository URI
Except where otherwise noted, this item's license is described as
File(s)![Thumbnail Image]()
Name
Emil_Tochev___UoN_Thesis___Corrections.pdf
Type
Full-text
Description
Examined
Size
6.48 MB
Format
Adobe PDF
Checksum (MD5)
cf88cc614f43a741d6e513ee5b2aadb2