2023 ASEE Annual Conference & Exposition

Challenges in Designing Complex Engineering Problems to Meet ABET Outcome 1

The first of the seven ABET outcomes is stated as “an ability to identify, formulate, and solve complex engineering problems by applying principles of engineering, science, and mathematics.” It further goes on to say that a “complex” engineering problem must meet one or more criteria such as having multiple solutions, no obvious or unique solution, Include many sub problems, involve multiple disciplines etc. While designing problems that meet the above criteria is achievable in capstone or project courses, such is not so in early freshman and sophomore courses. For this reason, an examination of collected student work reveals serious noncompliance with the requirements of Outcome 1. The collected work is often ordinary homework used as backfill. Examples include rudimentary coding, circuit analysis, signal processing, computer engineering, control and virtually all across the board. None even come close to passing the test of being a complex engineering problem. Although there is abundant literature on how to assess Outcome 1, there are very little examples of work that qualifies as such so others could benefit from. In this work, we present two examples of work that meet and surpass the definition of a complex engineering problem. The proposed problems have enough depth to even qualify as graduate projects and yet are understandable to any engineering freshman. The ideas are carried out in a first semester sophomore EE/CPE course dedicate to Matlab programming. The course currently running in Fall of 2022, has approximately 75 students in three parallel sections including both EE and CPE students. The corresponding outcome assessment is also presented.

Affectionately called Project Skittles, it began 4 years ago. The goal from the beginning has been identifying and segmenting skittles based on their color/flavor. This is truly a complex problem with no unique solution, or even any solution at all. Every “red” skittle is a different shade of red, has random highlights and may be randomly angled. Students need practically implement a Machine Learning algorithm, although they may not know it, to partition the RGB plane. Over the years, the scale of the problem has increased from a handful of skittles to a big jar. The objectives have become more ambitious as well. In its current incarnation, students are not only asked to segment the skittles by color, they need to provide a count of each flavor as well. The complexity of the problem ensures that no two student can possibly come up with the same count or the same color separation, and yet both could be right. Project is run in 3-student groups who assign tasks, milestones, and goals in Microsoft Teams. A detailed rubric is provided and is used for grading. Lastly, to test the robustness of the code, groups are asked to run the code on something other than skittles. One can’t but marvel at their imaginations. From sports jerseys, to wedding cakes and Christmas trees, some codes work extremely well but other break down. This is an opportunity to point out the importance of writing algorithms that are robust and do not work well on just one data set. Detailed metrics, scores and performance indicators are collected and analyzed. Invariably, some groups meet performance indicators that defy their limited knowledge. What has made this project popular year after year is its sheer simplicity and sophistication at the same time. It can be explained to anyone from kindergarten to college.

The same concept has been packaged as a project for satellite images of forest fires. In addition to the fires itself, students were asked to map six different land covers, from water, to land, green pastures etc. One step up from this was carried out last year using UC Merced satellite image database. The images were manually extracted from large images from the USGS National Map Urban Area Imagery collection for various urban areas around the country. The pixel resolution of this public domain imagery is 1 foot. It consists of 21 land cover types each represented by 100 picture frames. The idea is the same but more ambitious as each cover type is represented not by just one picture or color but by a complex scene. Using the same color separation approach students were fairly successful in achieving land cover separation but the problem was clearly above their aptitude as a sophomore. The good news is that it makes an appropriate follow up for capstone two years later.

The objective of this paper is to point out the difficulty of meeting the strict definition of ABET Outcome 1 and in the process present qualifying case studies that have actually been run in class many times.