Six Sigma Black Belt Project Examples

The project "Characterization of Turbofan Engine Production" was carried out as a part fulfillment for the Arizona State University (ASU) Six Sigma Black Belt Certificate program and to also assist Honeywell Engineers, Systems, and Services (HES), located in Phoenix.

We used the DMAIC approach based on the Six Sigma framework to accomplish this project. The project goals and time frame were defined in the Define phase. The goal of the project was to develop appropriate tools to improve the trending system currently being used in Honeywell that would point to a developing problem in the engine at early stage, before it leads to a major problem after some engines are shipped. All the data was available in the Honeywell PETS database which was used for analysis. The analysis was primarily done on the family TFE XXX engines to understand the working, failure modes and process variation so that it would help us in characterizing the new family engines. We were not involved in the improve phase of the approach because a new version of the engine namely AS900 is already in its initial phase of production. Process capability analysis of new engines showed that the process is operating at Six Sigma level. In the control phase of the project we designed Individual, Multivariate, and CUSUM control charts to control the production of new family of engines. Using these charts, the Honeywell AS907 Performance & Operability can trend the production of the engines in an advanced way than before. These charts would point to a developing problem quickly and reduce the number of engines required to determine any developing problems. It was estimated that incorporating these charts to monitor engine production would bring in a savings of approximately one half million dollars per annum.

Free Arts of Arizona is a non-profit organization dedicated to providing the healing effects of the creative arts to abused, neglected and homeless children by organizing events. For organizing these events, resources in terms of time and money are required. Free Arts organizes various fund raising events to generate funds. Additionally, these events help Free Arts in collecting valuable information about the donors and volunteers. The information about the donors is stored in a database at Free Arts of Arizona. The project to analyze data collected by Free Arts was completed in Six Sigma framework using DMAIC methodology. The scope of project was limited to providing recommendations for increasing the donations received by Free Arts of Arizona.

In the Define stage, the goals along with the expected benefits, of the project were identified jointly with the sponsor. The primary goal of the project was to analyze the data and provide recommendations which could increase the amount of donations. Also the project team and deadlines were decided during this stage of the project.

In the Measure state, the process of getting donations and organizing events to generate donations was understood by using a SIPOC diagram. The data already existed in the database, and this data was cleaned and retrieved from various tables by writing VB codes and queries. The predictor variables were then selected from the extracted data based on the domain knowledge and on the graphs of "Donation Amount" vs the predictor variables. Some of the predictor variables had too many levels, making statistical analysis difficult. The collapsing of the levels was done by combining levels which occurred less frequently into a new category, 'Other or miscellaneous'. Finally, a dataset was prepared for statistical analysis.

In the Analysis stage, the profile of donors and non-donors was analyzed using three approaches. In the first approach, an attempt to classify donors and non-donors was made, based on their profile using binary response. In the second approach, a model that could be used to predict the amount of donation was built. The first two approaches were not successful because of insufficient information about non-donors and because of high multi-colinearity among predictor variables. To deal with multi-colinearity, principal component regression was used; however, the best model did not explain even 25% variability in the predictors. In the third approach, the donation amount was discretized [sic] and a classification tree approach was used to clarify donors in 5 classes. The classification trees were built in R-Software and the best tree classified 57% data correctly. The tree also gave easy-to-understand paths that could potentially lead to high donation and that could be used for generating more funds. All the paths of the classification tree were analyzed to find ways that could help in increasing donations. The classification tree used 6 predictor variables for generating splits using gini measure. All these predictor variables were then analyzed in detail to see the contribution of different levels.

In the improvement stage, all the recommendations based on the classification tree and the graphical analysis of predictor variables were compiled and presented to Free Arts. Some important recommendations were:

- To send proposals to businesses and other organizations
- To follow up with the list of potential donors identified and given to Free Arts
- To approach individual donors who donate through United Way
- To invite more people to attend campaigns organized by Free Arts
- To add report generating feature in the database to generate reports directly from database that helps in decision making.

This project is implementation of Six Sigma methodology at Aerospace Division of Honeywell, Phoenix, AZ, USA. The objective of this project is to improve existent quality control system for aircraft engines manufacturing which would become capable of detecting potential problematic or underperforming engines at the above-mentioned facility. Every engine is tested for customer C.T.Q. (critical to quality). Additionally, every engine is tested for more than twenty quality specifications. Currently, Individual-Moving Rage (I-MR) control charts are used to monitor all of those variables separately. It is observed that although every engine passes quality specifications, some engines are not performing as well as other engines of same series. It leads to breakdown and underperformance of the engine, causing inconvenience to customer and warranty cost to Honeywell.

Six Sigma problem-solving methodology--DMAIC (Define, Measure, Analyze, Improve, and Control) has been used by us to improve the current quality system. We have the engine test results from 1996 to April 2003, which are used for analysis. The root cause analysis showed that the variables are correlated. The univariate control chart (I-MR) is effective for independent variables and not for correlated variables. Decusuib was taken to use Multivariate SPC - Hoteelling T2 Control Chart for monitoring the quality characteristics. Hotelling T2 Control Chart is effective only for 10 or few variables. Principal Component Analysis is used to reduce the dimensionality of the problem. The information with 20-plus variables is transformed into 6-8 variables, which accounts for 70% variability. The Hotelling T2 Control Chart is used on principal components calculated from historic data to check whether the process was in control during data collection. The normal operating behavior estimates were obtained, which are used for calculating Principal Components and corresponding Hotelling T2 statistic for new observation. Hotelling T2 control limits are established for monitoring new observations.

The newly proposed multivariate SPC - Hotelling T2 Control Chart and Principal Component Analysis Quality Control System is able to detect the engines, which are approved by I-MR control charts, but are candidates for potential failure. Six Sigma methodologies helped in improving the current quality system. It would result in delivery of engines which perform at 100% level for their life with minimum maintenance. It would result in increased customer satisfaction and reduced warranty cost for Honeywell.

In a highly competitive new manufacturing environment, customers demand increasingly shorter lead-time to reduce expenses. The demand pattern is also changing and becoming more unpredictable. The facility has been trying to solve this problem partially by keeping material on stock to reduce the risk of lost opportunity. But high WIP level and inventory period affects bottom line and creates other difficulties for smooth operation. Long WIP storage time could be observed between the Extrusion presses and the Tube draw benches in plant 2. Tracking the processing dates on route cards helped to reasonably quantify the time, the extruded profiles spend on the floor between these two processes. The analysis of this data also helped to identify long waiting time associated with a specific product codenamed 'XXOl'. This led to splitting the waiting time between extrusion and drawing processes to a special cause attributed to 'X:XO!' and a common cause across products. Work completed so far was focused on solving the special cause associated with 'XXOl'. Interviewing schedulers, quality control inspectors and examining available data about 'XXOl' revealed that the minimum yield strength required to start drawing operation couldn't be achieved immediately after extrusion. However, the yield strength can be improved by keeping the material on the floor for several days. The facility has been following the floor-age practice for 'XXOl' resulting in very high WIP storage period.

A designed experiment was conducted to understand the impact of different factors on yield strength during extrusion. A two-hole die was used to extrude aluminum billets of the same batch under varying levels of temperature and extrusion speed. The results of experimental runs revealed significant 'within-run' variability between the two profiles extruded from the same billets (run). Temperature had high influence on the profile extruded through one die hole, but was surprisingly found insignificant on the other profile from the same die. This led to the conclusion that the levels of temperature were different for the two profiles within the same experimental run. Further work is held up due to scheduling issues and machine breakdown, but results observed so far are expected to be beneficial in identifying and eliminating the route cause of variation between the two profiles. After reducing the variation, a response surface experiment could be conducted to find the region of maximum yield, which in turn will obviate the requirement of floor ageing. This will result in reducing WIP between extrusion and the tube drawing process.