Requirement Definition: Write the Test Specification First, Then Discuss the Proposal
1. Define the inspection types: Categorize them into four types—surface defects, dimensional measurement, character recognition, and positioning guidance—as the technical approaches vary significantly across different scenarios.
2. Quantify Core Metrics: Refine the objective of “identifying defective products” into specific, acceptance-testable criteria, such as “detect scratches ≥0.3 mm in length with a false negative rate ≤0.5%.”
3. Define Boundary Conditions: Specify the maximum production line cycle time, on-site vibration and lighting conditions, and integration requirements with PLC and MES systems.
Core Hardware Selection: Say No to Specification Inflating


1. Industrial Cameras: Calculate the required resolution using the formula “minimum feature size = 3–5 pixels”; the frame rate must match the production line cycle time; black-and-white cameras are preferred for dimensional measurement applications.
2. Lenses: Standard fixed-focus lenses are suitable for conventional workstations, while telecentric lenses are used for high-precision dimensional measurement to eliminate perspective distortion.
3. Lighting: Over 60% of stability issues in projects are caused by improper lighting. For reflective workpieces, diffuse dome lights are preferred; for low-contrast defects, custom-wavelength light sources are required.
Vendor Selection: Prioritize the Ability to Implement Solutions in Practice

1. Prioritize in-house development: Give preference to vendors with in-house control over algorithms and core hardware, and avoid integrators that rely entirely on outsourced components and assembly to minimize buck-passing when on-site issues arise.
2. Verification through real-world case studies: Require vendors to provide production lines that closely match your specific operating conditions, rather than a generic wall of brand case studies.
3. Assess iterative capabilities: Confirm that the vendor can provide a low-threshold data annotation and training platform that enables production line engineers to independently iterate on models without relying on on-site vendor support.
Closed-Loop Cost and Acceptance: Building Certainty into the Contract
1. Understanding the Cost Structure: Hardware accounts for only 15% to 30% of the total investment; software algorithms, on-site integration and debugging, and quality assurance iterations are the core value of the solution.
2. Fixed Acceptance Criteria: Agree in advance on the statistical criteria for miss-detection and detection rates (at the pixel or image level), IOU determination thresholds, and rules for sealing blind test samples to avoid vague clauses such as “accuracy ≥ 99%.”
3. Phased Acceptance: First, conduct blind testing on sealed samples; then, perform shadow operations after deployment and compare results with manual testing; finally, conduct closed-loop operations with small batches. The final payment should be made only after all criteria have been met.
You may also be interested in the following information
Let’s help you to find the right solution for your project!