Insights

Table 1: Repair Service Data Example

➤

Supplemental Data: Supplementary information related to production, such as product and manufacturing data. Sources mainly include internal development, engineering, and production departments, as well as external suppliers and distributors. The data mainly consists of four categories: reference product data, product design data, production-related data, and marketing data.

Table 2: Production-Related Data Example

In addition to the classifications above, data can also be structured in three ways, each answering a different question:

➤

Product Serial Number (Case Traceability): Tracks the complete lifecycle of a single product, from shipment to repair request, answering the question, “Why did this unit fail?”

➤

Production Batch (Cross-Sectional Comparison): Compares the concentration of failures among products from the same production period, answering the question, “Is there a problem with this batch?”

➤

Month (Time-Series Analysis): Observes overall trends over time, answering the question, “Is the situation getting better or worse?”

This diagram divides the path from data collection to application into four layers: supplemental data comes from systems such as SFC, ERP, and databases; the operational layer consists of daily records such as spare parts, workstations, and contracts; the analysis layer breaks these records down into interpretable metrics; and the decision layer converts the analysis results into concrete improvement actions.

The analysis layer at the top of this architecture is precisely the scope of quality analysis. Specifically, quality analysis can be understood as a pyramid that progressively narrows from macro trends across the entire lifecycle to the root cause of a single component:

Figure 2: Overview of Quality Analysis Pyramid Level 1–4 Definitions

Batch comparisons using AFR, responsibility attribution in defect determination, and field-level analysis in in-depth FA all require supplemental data to distinguish among design, manufacturing, and supplier causes. This is the data foundation behind the progressive narrowing from Level 1 through Level 4.

CFR (Cumulative Failure Rate) is calculated by dividing the number of warranty claims by the number of units shipped and accumulating the results month by month to form a trend curve. It is used to compare design quality differences among different PNs.

Figure 3: CFR Lifecycle-Wide Quality Trend Monitoring

AFR (Annualized Failure Rate) is derived using MTBF and the survival function, allowing products from different batches and with different shipment volumes to be compared on the same basis for manufacturing quality. The survival function R(t)=e^(-t/MTBF) essentially assumes that the product is in a random failure period where h(t) is approximately constant; CFR, in contrast, is a raw empirical rate without model assumptions. Used together, one shows trends while the other enables cross-sectional comparison.

Figure 4: Longitudinal Comparison of Batch AFR

The purpose of defect determination at the third level of the pyramid is to first clarify which party should be responsible for the returned item before conducting an in-depth root-cause analysis, thereby avoiding misallocation of subsequent analysis resources. Common classification methods include:

Figure 5: Defect Determination Responsibility Attribution Matrix

Proper classification must come first to ensure that subsequent analysis addresses the right cause. If a CID (Customer-Induced Cause) is mistakenly classified as a PID (Process-Induced Cause) and the production line is investigated in depth, it not only wastes resources but may also lead to unnecessary process adjustments.

The final level of the pyramid, in-depth FA (Failure Analysis), analyzes data layer by layer across dimensions such as repair codes, failure causes, suppliers, and component PNs to identify potential risks across three major dimensions: design, manufacturing, and warranty. Common tools such as fishbone diagrams (systematically listing possible causes) and Pareto charts (ranking causes by occurrence rate and prioritizing key causes) can locate the root of a problem from different perspectives. The dashboard shows “where the anomaly is,” while this toolkit answers “why the anomaly occurred.”

Figure 6: Field-Level Analysis Path for In-Depth FA

➤

VID (Vendor-Induced Cause): Feed findings back to procurement and supplier quality teams, including the supplier of defective parts, PN, and batch number, as a basis for supplier audits and incoming inspection adjustments.

➤

CID (Customer-Induced Cause): Feed findings back to the design team and customer service/training departments, including common misuse scenarios, to optimize user manuals, error-proofing designs, or training priorities.

For the feedback loop to truly work, the key is to structure the analysis conclusions. Reports should be generated using standardized fields (responsibility category, root cause, recommended action, and responsible department) so that each root cause can be tracked to determine whether it was actually adopted and what results it achieved.

The endpoint of the quality analysis pyramid should not be merely a report, but the reintegration of its findings into the design and manufacturing processes of next-generation products:

Figure 7: Data-Driven Continuous Improvement Loop