In the course of my work as an engineer specializing in powertrain manufacturing, I encountered a critical quality challenge: the persistently high rejection rate of engine cylinder block castings due to surface defects. The problem was particularly acute for the CA4GA1 engine, a 1.3L gasoline engine developed for economy cars. The cylinder block castings produced in our foundry suffered from a surface defect scrap rate of 13.26% in 2012, compared with an average of approximately 1.8% for three-cylinder engine blocks. This not only increased production costs but also constrained capacity and delayed deliveries to customers. Determined to find a systematic solution, I led a Six Sigma project following the DMAIC methodology to identify root causes and implement effective improvements. This article describes the entire journey, including the tools used, the statistical analyses performed, and the significant results achieved.
Introduction to the Sand Foundry Defects Problem
Engine cylinder blocks are fundamental components of an internal combustion engine. They house the cylinders, support the crankshaft and pistons, and provide passageways for cooling and lubrication. Therefore, their quality directly affects engine performance and reliability. In our foundry, the CA4GA1 cylinder block was produced using sand casting, and the major categories of sand foundry defects were identified as sand adhesion, blowholes, and sand holes. These defects were strongly influenced by process parameters such as pouring speed, core loss on ignition (LOI), and sand mold hardness. The high scrap rate not only wasted material and energy but also increased rework labor and reduced overall productivity. To address this issue, I adopted the Six Sigma management approach, which provided a structured framework for problem solving.
Six Sigma Methodology and DMAIC
Six Sigma is a data-driven methodology that aims to reduce variation and defects in processes. The core framework is DMAIC: Define, Measure, Analyze, Improve, and Control. Each phase has specific objectives and tools. In the Define phase, we clarify the problem and set project goals. In the Measure phase, we establish a baseline and assess the measurement system. In the Analyze phase, we identify the significant factors that affect the output. In the Improve phase, we optimize the process parameters to reduce defects. Finally, in the Control phase, we implement monitoring mechanisms to sustain the improvements. Throughout this project, I used a wide range of statistical tools, including Pareto analysis, measurement system analysis (MSA), process capability analysis, cause-and-effect matrices, failure mode and effects analysis (FMEA), hypothesis testing, design of experiments (DOE), regression analysis, and response surface methodology. The remainder of this article details each phase of the project.
Define Phase
At the beginning, I established the project scope and objectives based on the company’s strategic focus on quality improvement and cost reduction. The main customer feedback from the downstream machining department indicated that sand foundry defects on the CA4GA1 cylinder block caused frequent stoppages and excessive tool wear. Through a Pareto analysis of defect categories, I identified that surface defects were the dominant contributor to the overall scrap rate. Among these surface defects, sand adhesion, blowholes, and sand holes accounted for the majority. Therefore, I defined the output variable Y as the surface defect scrap rate of the CA4GA1 cylinder block castings. The project goal was to reduce the scrap rate from 13.26% to below 4%. I also estimated the potential cost savings: with an annual production of 150,000 castings and a cost of 156 RMB per casting, reducing the scrap rate by 9.26 percentage points would yield direct savings of over 2.1 million RMB per year. Soft benefits included increased production capacity, reduced downstream rework, and improved customer satisfaction. The project team included process engineers, quality engineers, and operators, all working together under my coordination. We established a timeline for each phase and set up regular review meetings to track progress.
Measure Phase
Measurement System Analysis (MSA)
Before analyzing data, I needed to ensure that our measurement systems were reliable. For the surface defect count, which is attribute data, I conducted a gauge repeatability and reproducibility (R&R) study using 30 samples, of which 20 had known defects. Three appraisers evaluated each sample twice. The results showed an overall effectiveness of 93.33%, which exceeded the 90% threshold, confirming that the attribute measurement system was trustworthy.
For the key continuous variables—pouring speed (Y1), core LOI (Y2), and mold hardness (Y3)—I performed continuous MSAs. For pouring speed, three inspectors timed the pouring duration for 10 castings twice. The %P/TV was 26.98% (less than 30%), and the number of distinct categories was 10 (greater than 5), indicating an acceptable measurement system. For core LOI, the %P/TV was 20.41% and the number of distinct categories was 12, also acceptable. For mold hardness, the %P/TV was 16.06% with 15 distinct categories, again satisfactory. These studies confirmed that the data collected for analysis accurately reflected the true process performance.
Process Capability Analysis
I then assessed the current process capability for each output. The surface defect scrap rate over 16 weeks exhibited a stability but a very low capability, with a calculated sigma level of only 2.1. For pouring speed, the capability index Cpk was 0.88, which was below the minimum acceptable value of 1.33. The core LOI had a Cpk of 0.97. The mold hardness had a Cpk of 1.04. All these values indicated that the processes were not capable of meeting the required specifications consistently. This was the baseline from which improvements had to be made.
Cause-and-Effect Matrix
To systematically identify potential root causes, I constructed a cause-and-effect (C&E) matrix linking the process inputs to the three outputs Y1, Y2, and Y3. The matrix is shown below:
| Process Input | Importance for Y1 (pour speed) | Importance for Y2 (core LOI) | Importance for Y3 (mold hardness) | Total Score |
|---|---|---|---|---|
| Raw sand moisture | 0 | 9 | 0 | 81 |
| Resin addition | 0 | 9 | 0 | 81 |
| Holding time | 9 | 0 | 0 | 90 |
| Pouring temperature | 9 | 0 | 0 | 90 |
| Runner position | 9 | 0 | 0 | 90 |
| Drying temperature | 0 | 9 | 0 | 81 |
| Drying time | 0 | 9 | 0 | 81 |
| Bentonite addition | 0 | 0 | 9 | 72 |
| Water addition | 0 | 0 | 9 | 72 |
| Mixing time | 0 | 0 | 9 | 72 |
| Mold strength | 0 | 0 | 9 | 72 |
| Compactability | 0 | 0 | 9 | 72 |
Using a Pareto analysis on the total scores, I prioritized the most important inputs. Some of these were addressed through quick improvements, while others required further investigation.
Failure Mode and Effects Analysis (FMEA)
I facilitated a team FMEA to identify potential failure modes for each significant process step. The initial FMEA revealed several high-risk items with RPN (Risk Priority Number) values exceeding 100, including runner position, pouring temperature, raw sand moisture, resin addition, curing time, stirring time, penetration depth, drying time, drying temperature, door opening condition, bentonite addition, water addition, mixing time, mold strength, compactability, and impact pressure. For each of these, I proposed corrective actions. Some could be implemented immediately as quick improvements, such as installing automatic water addition devices, setting standard samples for penetration depth, adding time relays for curing, and establishing a simple lever for the oven door. These quick improvements were implemented by the team members with specific responsibilities and deadlines.
After implementing the quick improvements, I repeated the FMEA to reassess the risks. Many RPN values decreased significantly. However, seven factors still remained relatively important: runner position (X1), pouring operator variability (X2), pouring temperature (X3), drying temperature (X4), drying time (X5), mold strength (X6), and compactability (X7). These were the candidates for detailed analysis in the next phase.
Analyze Phase
The goal of the Analyze phase was to statistically validate which of these potential X factors truly had a significant impact on the sand foundry defects. I began with a multi-vari analysis on the pouring speed Y1.
Analysis of Pouring Speed Y1
I collected pouring speed data from 30 morning shifts and 30 afternoon shifts. A one-way ANOVA yielded a p-value of 0.047, which was slightly above the 0.05 significance level, indicating that the shift change did not have a statistically significant effect on pouring speed. I then tested the effects of runner position, operator variability, and pouring temperature. The results are summarized below:
| Factor | p-value | Conclusion |
|---|---|---|
| Shift | 0.407 | Not significant |
| Runner position (X1) | 0.000 | Significant |
| Operator variability (X2) | 0.934 | Not significant |
| Pouring temperature (X3) | 0.001 | Significant |
Thus, runner position and pouring temperature were the significant factors affecting pouring speed, while operator variability was not.
Analysis of Core LOI Y2
For core LOI, I designed a two-factor two-level full factorial experiment with three center points and one replicate to test the effects of drying temperature (X4) and drying time (X5). The factor levels were 180°C and 220°C for drying temperature, and 1.5 and 2.5 hours for drying time. The DOE data were analyzed using Minitab. The main effects plot showed a nonlinear relationship, meaning curvature existed. The ANOVA results showed that both drying temperature and drying time, as well as their interaction, had p-values less than 0.05:
| Term | Effect | Coefficient | p-value |
|---|---|---|---|
| Constant | 2.4425 | 0.000 | |
| Drying temperature | 0.0349 | 0.0174 | 0.000 |
| Drying time | -0.1050 | -0.0524 | 0.000 |
| Temperature × Time | 0.0749 | 0.0374 | 0.000 |
The R² was 99.65%, confirming the model’s excellent fit. Both factors and their interaction were significant, and the curvature indicated the need for a response surface approach later.
Analysis of Mold Hardness Y3
Similarly, I performed a factorial experiment for mold strength (X6) and compactability (X7). The levels were 1.6 and 2.4 for mold strength, and 28 and 42 for compactability. The main effects plot revealed curvature. The ANOVA results showed significant effects for both factors and their interaction, with p-values below 0.05. The R² was 98.77%.
| Term | Effect | Coefficient | p-value |
|---|---|---|---|
| Constant | 85.874 | 0.000 | |
| Mold strength | 3.749 | 1.874 | 0.000 |
| Compactability | -2.250 | -1.125 | 0.002 |
| Strength × Compactability | -1.250 | -0.625 | 0.026 |
Thus, both factors and their interaction significantly affect mold hardness. The curvature again suggested the need for response surface optimization.
Improve Phase
Based on the analysis, I knew which parameters had to be adjusted. I now used regression analysis and response surface methodology to find the optimal settings.
Regression Analysis for Pouring Speed Y1
I collected data on pouring speed at various pouring temperatures. I fitted linear, quadratic, and cubic regression models. The following table compares them:
| Model | R² | Adjusted R² | S | Evaluation |
|---|---|---|---|---|
| Linear | 30.6% | 28.5% | 1.196 | Poor |
| Quadratic | 83.6% | 82.5% | 0.591 | Best |
| Cubic | 83.7% | 82.2% | 0.597 | Good |
The quadratic model was chosen because it had the lowest S and a high adjusted R². The resulting regression equation was:
$$Y_1 = -22340 + 32.37 X_3 – 0.01171 X_3^2$$
By analyzing the fitted curve and the 95% prediction interval, I determined that the optimal pouring temperature range was 1375°C to 1390°C. Within this range, the pouring speed fell into the desired 22–26 seconds, minimizing sand adhesion defects.
Response Surface Optimization for Core LOI Y2
To account for the curvature observed in the factorial analysis, I applied response surface methodology (RSM) using a central composite design. The data for core LOI (Y2) with drying temperature (X4) and drying time (X5) are given in the following table:
| Run Order | Drying Temp (°C) | Drying Time (h) | Core LOI (%) |
|---|---|---|---|
| 1 | 180 | 1.5 | 2.51 |
| 2 | 220 | 1.5 | 2.47 |
| 3 | 180 | 2.5 | 2.33 |
| 4 | 220 | 2.5 | 2.45 |
| 5 | 171.7 | 2.0 | 2.36 |
| 6 | 228.3 | 2.0 | 2.51 |
| 7 | 200 | 1.29 | 2.52 |
| 8 | 200 | 2.71 | 2.36 |
| 9 | 200 | 2.0 | 2.31 |
| 10 | 200 | 2.0 | 2.30 |
| 11 | 200 | 2.0 | 2.30 |
| 12 | 200 | 2.0 | 2.31 |
| 13 | 200 | 2.0 | 2.31 |
The response surface regression yielded the following equation:
$$Y_2 = 2.306 + 0.03652 X_4 – 0.05328 X_5 + 0.06512 X_4^2 + 0.06763 X_5^2 + 0.04 X_4 X_5$$
All coefficients were significant (p < 0.05). Since core LOI is a “smaller is better” characteristic, I used the response optimizer to find the minimum. The optimum was achieved at a drying temperature of approximately 191°C and a drying time of about 2.26 hours, yielding a predicted core LOI of 2.28%. This setting would reduce gas evolution and blowhole defects.
Response Surface Optimization for Mold Hardness Y3
I performed a similar response surface analysis for mold hardness (Y3) as a function of mold strength (X6) and compactability (X7). The regression equation was:
$$Y_3 = 93.6 + 1.8839 X_6 – 0.9268 X_7 – 3.4250 X_6^2 – 4.4250 X_7^2 – 0.5 X_6 X_7$$
Mold hardness should be maximized to prevent sand holes. The response optimizer indicated that the maximum hardness of about 93.9 was obtained when mold strength was set around 2.1 and compactability around 34.1. Thus, the optimal ranges were: mold strength 2.0–2.2 and compactability 32–36.
Summary of Optimal Settings
The critical parameter settings from the analysis are summarized below:
| Parameter | Optimal Value / Range |
|---|---|
| Pouring temperature | 1375°C – 1390°C |
| Drying temperature | 185°C – 195°C |
| Drying time | 2.2 – 2.4 hours |
| Mold strength | 2.0 – 2.2 |
| Compactability | 32 – 36 |
| Pouring speed (target) | 22 – 26 seconds |
| Core LOI (target) | 2.24 – 2.33% |
| Mold hardness (target) | 92 – 96 |
After implementing these improved parameter settings, the surface defect scrap rate dropped significantly. By the end of July, the scrap rate had fallen from 7.49% to 4.15%.
Control Phase
To sustain the gains, I developed a comprehensive control plan. The plan includes specific monitoring methods, sampling sizes, frequencies, and responsible departments for each critical X and Y variable. The key elements are shown below:
| Item | Variable | Specification | Measurement Method | Tool | Sample | Frequency | Control Chart |
|---|---|---|---|---|---|---|---|
| 1 | Runner position | Optimum design | Review | Engineering drawing | Monthly | Per month | None |
| 2 | Pouring temperature | 1375–1390°C | Measurement | Thermocouple | Every ladle | Each ladle | Individuals chart |
| 3 | Drying temperature | 185–195°C | Program control | Digital display | 100% | Continuous | None |
| 4 | Drying time | 2.2–2.4 h | Program control | Digital display | 100% | Continuous | None |
| 5 | Mold strength | 2.0–2.2 | Measurement | Hydraulic strength tester | 4 samples | Per shift | Xbar-R |
| 6 | Compactability | 32–36 | Measurement | Compactability tester | 4 samples | Per shift | Xbar-R |
| 7 | Pouring speed | 22–26 s | Measurement | Stopwatch | 5 samples | Daily | Xbar-R |
| 8 | Core LOI | 2.24–2.33% | Measurement | Electronic balance | 5 samples | Daily | Xbar-R |
| 9 | Mold hardness | 92–96 | Measurement | Hardness tester | 5 samples | Daily | Xbar-R |
I also updated and standardized the relevant process documents. The following files were revised:
| Document Name | Document Number | Effective Date | Change Description |
|---|---|---|---|
| Cylinder Block Casting Process Drawing | 4GA14004000-2211-2 | 2013-06-17 | Improved runner position |
| Pouring Work Instruction | TN-3215-1 | 2013-08-19 | New pouring temperature range |
| Core Drying Work Instruction | TN-3207-1 | 2013-08-19 | New drying temperature and time ranges |
| Cylinder Block Sand Inspection Card | TN-2502-55-1 | 2013-08-19 | New mold strength and compactability limits |
| Sand Drying Work Instruction | TN-3201-1 | 2013-05-15 | Random inspection timing |
| Core Coating Work Instruction | TN-3206-3 | 2013-05-18 | Continuous stirring required |
| Hot-box Resin Mixing Instruction | TN-3201-2 | 2013-05-18 | Time relay for curing |
| Cylinder Block Sand Mixing Instruction | TN-3209-3 | 2013-08-18 | Automatic water addition and mixing improvement |
| Hot-box Core Making Instruction | TN-3203-6 | 2013-05-18 | Resin dosing cylinder |
For ongoing process monitoring, I implemented statistical process control (SPC) using Xbar-R charts for the continuous variables. As an example, the Xbar-R chart for core LOI (shown below) demonstrated that the process remained stable after the improvements. All points were within the control limits, and no special cause patterns were observed.
Using the data collected from 20 August to 30 August 2013, I performed a Poisson capability analysis on the surface defect count. The average DPU was 0.0322, with the U chart showing the process was in control. A two-proportion test comparing the defect rate before and after the improvements gave a p-value of 0.000, confirming a statistically significant reduction in sand foundry defects.
Results and Benefits
By the end of September 2013, the surface defect scrap rate for CA4GA1 cylinder block castings had declined to 2.02%. This was well below the initial target of 4%, representing a target achievement rate of 198%. The hard financial savings were calculated as:
$$150{,}000 \times (13.26\% – 2.02\%) \times 156 = 2{,}630{,}160 \text{ RMB}$$
In addition to these direct savings, the reduction in sand foundry defects led to a 12% increase in cylinder block production capacity, reduced rework and energy consumption in the cleaning and machining departments, shortened delivery times, and higher customer satisfaction.
Conclusions
This project demonstrated the power of Six Sigma methodology in tackling a complex manufacturing problem. By systematically following the DMAIC phases, I was able to move from a vague understanding of “high scrap rate” to a precise identification of the critical process parameters affecting sand foundry defects. The combination of statistical tools allowed me not only to find the root causes but also to optimize the process settings scientifically. The improvements were sustained through standardized documents and SPC monitoring. The success of this project reaffirmed my belief that data-driven problem solving, when combined with a structured methodology and a committed team, can deliver exceptional results. Six Sigma is not just a set of tools; it is a mindset that encourages continuous improvement and excellence. I hope this article provides a useful reference for others facing similar challenges in foundry operations or other manufacturing processes.