ICT Probe-Induced Indentations on PCB Pads - IPIP²
Description
Bed-of-nails fixtures used in In-Circuit Testing can contain hundreds of spring-loaded probes and withstand thousands of inspection cycles without maintenance. During testing, each probe may apply forces of up to 4 N on the PCB surface—enough to pierce protective coatings and ensure reliable electrical contact with the conductive pad. However, repeated mechanical loading, poor probe alignment, or mechanical wear can damage the internal spring mechanism, increasing the applied force on the test pads and producing deeper indentations. Over time, these indentations may compromise the solder mask and/or expose underlying copper traces, potentially affecting the board’s functionality and long-term reliability. This dataset provides a collection of PCB pad images acquired via microscopy, covering multiple combinations of pad coatings and probe types. It is intended to support the development of optical Nondestructive Evaluation techniques capable of segmenting and classifying probe-induced indentations as acceptable (OK) or critical (NOK). Contents: -1853 pad images with probe-induced indentations produced by two needle types, identified as 1025 and 1012. Each filename follows a consistent naming convention. For example, the file “1025_OSPAT4_C_PCB2.051_9.4_75” encodes key acquisition details: 1025 identifies the probe (needle) type, OSPAT4 indicates the pad coating, and the value immediately before “_75” corresponds to the force applied by the probe that produced the indentation on the pad surface. This force value can be used as ground truth for classification tasks. Based on it, developers may define a force threshold to separate OK from NOK. According to most probe manufacturers, the nominal force of a probe in good condition is below 4 N. -_masks_indentations: binarized masks with indentation regions segmented (ground truth) for training indentation segmentation/detection models. -_masks_pads: binarized masks with the pad region activated (white pixels), intended as ground truth for pad segmentation. -_indentation_features: Excel files (one per coating–needle pair) containing 500+ variables describing geometric and photometric properties of each indentation. -GetAllFeatures.hdvp: procedure used to extract the variables stored in the _indentation_features spreadsheets.~ How the dataset can be used -The provided data enables the development of a complete computer-vision pipeline, typically involving: -Pad segmentation to isolate the pad region; -Indentation segmentation to localize probe marks within the pad; -Pad-level classification (OK/NOK) based on indentation appearance and/or extracted properties. Accordingly, the dataset supports training and testing: -pad segmentation models (from full pad images); -indentation segmentation/detection models (using indentation masks); -classification models that use either: cropped indentation images (common in deep-learning classifiers), or tabular indentation features (for machine-learning approaches).
Files
Institutions
- Universidade do Minho
Categories
Funders
- Missão Interface of the Recovery and Resilience Plan (PRR)Grant ID: 01/C05-i02/2022