Quality Inspection Data Collection: Design, Architecture, and Applications

1 Background

Every device that passes the official verification at ZhiZhuan undergoes comprehensive testing at a quality inspection station. While the testing line records results, the actions, hardware parameters, and time data generated during the inspection are not systematically utilized. These process data can reflect the entire workflow of quality inspection engineers, and we aim to capture them through technical means.

2 Quality Inspection Data Points

To better collect various data during the inspection process, we researched common industry data‑point designs and, based on the actual inspection site, formulated an operation data‑point scheme that fits the quality inspection workflow.

2.1 Data Point Design

The inspection line involves multiple terminals and platforms. Although each terminal reports to its own data‑point platform, differing standards and fragmented platforms prevent the formation of a structured execution data chain. Our solution places data points within the client based on its specific inspection scenarios. The data‑point structure follows a predefined schema, while the actual content is defined per scenario. All points are collected by a unified data‑point platform that stores logs, provides a data‑processing service, and offers real‑time BI query tools.

2.1.1 Data Point Structure

We modeled the inspection process and defined the data‑point elements as: user, action, timestamp, business parameters, and operating environment information.

2.1.2 Data Point Reporting

Reporting mechanism: scenario trigger → standard data extraction → data upload → fallback retry → coarse filtering → storage → big‑data cleaning.

2.2 Data Point Architecture

After each inspection node uploads its operation data, the flow and usage of this data within the overall architecture are shown below:

For storage, data is first written to the data‑point platform's business database; a CDC synchronization mechanism then replicates it to the big‑data side, where secondary processing and cleaning occur.

3 Data Point Applications

By analyzing the standardized data reported from data points and combining it with appropriate data standards and strategies, we can achieve digital management of the inspection line.

3.1 Scenario – Compliance Execution

In compliance judgment strategies, the actions of inspection engineers are analyzed in real time to determine whether they follow the standard SOP. For example, the "Mobile Phone Shell Appearance" inspection requires the engineer to rotate the device around multiple angles according to the SOP.

--《手机外壳外观》标准质检SOP
s1：屏幕息屏正面对着质检师，从上往下，依次检查是否有划痕、瑕疵等。
s2：从中框任意一个点开始，设备环绕一圈检查中框情况
s3：屏幕息屏背面对着质检师，从上往下，依次检查是否有划痕、瑕疵等。

This utilizes sensor data collected from the device during inspection.

The three‑dimensional (X, Y, Z) angle sensor data forms a line chart.

When the SOP requires a full‑circumference check of the "four sides of the frame," the X‑axis angle values follow a pattern such as 0°→90°→180°→‑180°→‑90°→0°, with similar trends on other axes. Based on these patterns we define compliance angle requirements.

Step

X‑axis

Y‑axis

Z‑axis

Duration

Screen Front

±60°

2s

Four‑Side Frame

±180°

±20°

3s

Phone Back Cover

±60°

±180°

1s

When the angle stays within the configured range for longer than the specified duration, the action is considered compliant.

Illustration of compliant angle zones (red area indicates compliance).

3.2 Scenario – Intelligent Strategy

Beyond real‑time analysis, a major application is secondary data processing. Leveraging big‑data cleaning, transformation, and mining, we have built a complete analysis system. In quality inspection, data‑driven algorithms break the limits of fixed‑threshold strategies, enabling self‑iterating intelligent strategies.

As shown, business data undergoes a processing loop that generates new data, forming a closed‑loop that drives strategy self‑iteration. Thus, inspection strategies evolve from static thresholds to intelligent, algorithm‑enhanced policies.

Diagram of intelligent strategy interaction.

3.3 Scenario – BI Dashboard

Based on the defined inspection behavior data points, after processing the data we can generate detailed, multi‑dimensional BI dashboards.

4 Conclusion

We are currently in the early stage of digitalizing quality inspection through data‑point exploration. Looking ahead, we will deepen digitalization, leveraging big‑data and algorithmic capabilities to broaden the scope of inspection digitization. The road is long, but continued effort will lead us forward.

5 References

https://zhuanlan.zhihu.com/p/577455883

https://zhuanlan.zhihu.com/p/665167127

About the author Tú Zhìwǔ, from ZhiZhuan Group – R&D Center – Fulfillment Business Platform – Quality Inspection Technology Department, responsible for backend development of ZhiZhuan's inspection system.