Insight / Blog
Sand 3D Printing Quality AI : Defect Detection Across 1,500 Layers
Sand 3D printing builds casting molds and cores by repeatedly recoating a level layer of sand and jetting a liquid binder. Thousands of recoating and printing cycles accumulate into one precision additive-manufacturing job.
A single run can continue for many hours. Small layer errors can accumulate and affect the quality of a large mold, making real-time inspection critical during printing.
Background : Monitoring Up to 15 Hours and 1,500 Layers
Company A, a sand 3D printer manufacturer, needed to manage defects during long printing cycles. Operators had to watch screens and process video for as long as 15 hours while monitoring approximately 1,500 layers.
Even with dedicated monitoring, fatigue can lead to missed defects. Recoating and printing errors may remain subtle during a run but create significant downstream quality costs. Operators needed a clear view of error type, location, and risk level.
Approach : Combined Video and Vibration Analysis
SURROMIND designed an AI quality-inspection pipeline that combines process video with equipment vibration data to reduce continuous monitoring load and improve decision consistency.
- Video and vibration analysis: combined process imagery with vibration patterns from recoating to identify anomalies and error types.
- Process-specific decisions: evaluated recoating and binder-printing stages separately for error presence and type.
- Location and risk visualization: displayed defect locations and categorized them as safe, caution, or high risk according to size and position.
- Equipment integration: connected trained inspection models with the machine interface so operators could review results and respond in the field.
Results : 0.99 Recall for Error Detection and Process-Level F1 Validation
| Metric | Previous monitoring | SURRO Inspection |
|---|---|---|
| Monitoring effort | Extended manual screen monitoring | Real-time AI monitoring reduces continuous observation load |
| Process error detection | Visual identification with a risk of missed decisions | Video and vibration based error detection with Recall 0.99 |
| Defect-type diagnosis | - | Recoating F1-score 0.97; printing F1-score 0.99 |
| Field response | Primarily post-run review | Error-location visualization and location-based risk classification |
| Inspection data | Target: 15,000 records | 22,660 records collected |
Operational Impact : Real-Time Response by Error Location and Risk
Improving the reliability of sand 3D printing requires both lower monitoring effort and more consistent decisions. SURRO Inspection was designed around those two operating needs.
The system monitored recoating and printing quality across approximately 1,500 layers. Operators could focus on the error types and risk locations identified by AI instead of watching every frame continuously.
The workflow also analyzed vibration patterns during recoating, adding a signal source for anomalies that were difficult to identify from images alone. When an error was detected, its location and risk level were displayed for operator review.
The project collected more inspection data than planned, validated strong error-detection and defect-type performance across both processes, and established a field-ready operating foundation for real-time quality monitoring.
