The objective of this project is to create numerous datasets with an anomaly detection device (Delphi) and improve the internal and external design of the current prototype.
Sponsored by: TDI Novus, Inc.
Team Members
Tingxue Gu Shareef Parkes Christian Piccioni ]Sebastian Rodriguez Traconis Yiyang Wang
Instructor: Charlie Purdum
Project Poster
Click on any image to enlarge.
Project Video
Project Summary
Overview
This fall our group was selected to work with TDI Novus through Penn State’s learning factory and department of Industrial Engineering. TDI designs forward-thinking AI and robotics solutions for Government and commercial clients. Our sponsor Adam Blenk and his team developed a prototype called Delphi that uses machine learning algorithms run through a raspberry pi to detect abnormalities in the behavior of machines.
Objectives
TDI gave us two primary objectives. First, they wanted to collect test data using Delphi on as many different applications as possible. TDI has limited access and resources to conduct testing, so Penn State’s expansive system of facilities and labs helped fulfill their need. Second, TDI asked us to use CAD software to redesign and optimize Delphi’s casing. Eventually we also explored the possibility of using a raspberry pi mini and further decreasing the size of Delphi.
Delphi records vibrational frequencies using an accelerometer during a 20-minute training period. Once the algorithm was trained, we induced an artificial fault on the machine and monitored Delphi to see if it would successfully detect the abnormality. Delphi has the capability to alert an operator when it determines a machine to be out of control. This simple action can save a company from costly disasters.
Approach
-Sponsor visited Penn State to demo the prototype and lay out objectives
-The team coordinated outreach to PSU Stakeholders (Office of Physical Plant, Fame Lab) to determine what machines we could test on
-Data collection done on 10 different machines at Penn State
-Data was validated by looking at output charts alongside our sponsor
-Data analysis was not in the scope of our project and will be left to the sponsor
-CAD modeling techniques used to redesign sensor mount and Delphi casing
-Materials for the new prototype were researched and procured by team members
-Engaged KCF technologies for guidance on casing design and waterproofing
-Casing model and sensor mount were 3d printed multiple times to ensure the best material available was used
Outcomes
Our team determined that Delphi performs best when monitoring machines that run with consistent and smooth vibrations for long periods of time. Delphi’s redesigned casing is XX% smaller than the initial prototype. This will increase its opportunities for application and our work with raspberry pi mini is encouraging for making Delphi even smaller and more versatile in the future. Our project gave TDI access to testing opportunities that they otherwise would not have been able to conduct without spending significant amounts of time and money to get access to material.
We are leaving TDI with a better understanding of Delphi’s strengths along with clear opportunities to make their product more robust so that it adds the most value possible for their customers.
–>