Conventional PCB fault detection are inefficient and error-prone, which tells us that automation is necessary. For this research, YOLOv5 is used onRaspberry Pi to detect sixcategories of PCB defects including MouseBites, Open Circuits,andShort Circuits in real-time. This system is offline capable and provides fast offline detection, accuracy, .and low power optimization. Its accuracyisconfirmedthroughprecision,recall,andmAPperformancemeasures.FindingsindicatethatAIdefectdetectionimproves defect detection significantly, and thus greatly increases PCB quality control while minimizing costs and reducing the ineffecient manual inspections
Introduction
The text discusses the importance of quality control in PCB (Printed Circuit Board) production, highlighting issues like microcracks, shorts, and soldering defects that reduce device reliability and increase servicing costs. Traditional manual inspections and electrical probes are prone to human error, especially with complex multi-layer PCBs. This has driven the need for more efficient automated fault detection methods.
Automated Optical Inspection (AOI) and AI-based techniques, particularly deep learning with Convolutional Neural Networks (CNNs), have gained traction. Among these, the YOLO (You Only Look Once) algorithm is favored for fast and accurate real-time fault detection. This study focuses on applying the latest YOLOv5 version on a Raspberry Pi platform to create an affordable, efficient, and offline real-time PCB defect detection system. It detects six defect types (Mouse Bites, Open Circuits, Short Circuits, Spurs, Misaligned Components, and Soldering issues) by processing uploaded PCB images, highlighting defects with bounding boxes, and operating efficiently despite Raspberry Pi’s limited processing power.
The literature review covers the evolution of deep learning in object and defect detection, emphasizing YOLO’s balance of speed and accuracy in various applications, including PCB inspection and assistive devices. Prior studies validate YOLOv5’s superior performance over previous versions.
The methodology details the system architecture: images are input to the YOLOv5 model on Raspberry Pi, which detects defects, displays results, and stores data for quality control. The YOLOv5 architecture uses improved bounding box prediction and loss functions to enhance localization accuracy. Due to Raspberry Pi’s limited hardware, model training is done on a powerful PC, then the model is optimized (e.g., converted to TensorFlow Lite) and deployed with hardware acceleration on Raspberry Pi. A camera module enables live PCB inspection.
Evaluation metrics like precision, recall, mean average precision (mAP), latency, and frames per second (FPS) assess system performance to ensure real-time effectiveness in practical PCB quality control.
Conclusion
Insummary,theYOLOv5defectinspectionsystemisperfectionforPCBmanufacturingqualitycontrolthroughtheassureddetectionofmultipledefectswithinasinglePCB.Identificationofdefectsofbothtypeandlocationbythesystemwillthusbeanenablingfactorformanufacturersinqualitycontrol,particularlyprioritizingrepairattempts,hencekeepingcostslowwhile improving better product reliability.
Above all, predictive accuracy that captures the level of defect severity is the final piece of the puzzle missing, which ismost needed to direct the production and also set priorities in addressing those issues which are great trouble for the production. In making the system foolproof and scalable at will to address industry needs, more optimizations to the system in the sense of affixing measures of confidence and thorough testing from real-world cases are required.
Sinceweintendtorefineandtestthesystem,weshallbeemphasizingincreasingitscapabilitytosupportmultipledesignsand manufacturing condition for PCB. It is intended to improve reliability elimination of spurious positives, ease PCB manufacturing, and facilitate effective error-free E-manufacturing. The benefit of technology to the entire is vastly enormous forconsumersand producersaliketowitnessimprovementsinproductqualityandthereductionoftheoverheadofproduction for the entire electronics sector.
References
[1] Phadnis,Rasika&Mishra,Jaya&Bendale,Shruti. (2018). Objects Talk -ObjectDetectionandPattern Tracking Using TensorFlow. 1216-1219. DOI:10.1109/ICICCT.2018.8473331.
[2] R.Parvadhavardhni,P.Santoshi,andA.M.Posonia,\"BlindNavigationSupportSystemusingRaspberryPi&YOLO,\"20232ndInternationalConferenceon Applied Artificial Intelligence and Computing (ICAAIC), Salem, India, 2023, pp. 1323-1329. DOI: 10.1109/ICAAIC56838.2023.10140484.
[3] Cai, Li, and Jingchuan Li. \"PCB Defect Detection System Based on Image Processing.\"Journal of Physics: Conference Series, Vol. 2383, No. 1, IOPPublishing,2022.
[4] Chen, Shiqiao, Xiqing Liang, and Wenneng Jiang. \"PCB Defect Detection Based on Image Processing and Improved YOLOV5.\"Journal of Physics:Conference Series, Vol. 2562, No. 1, IOP Publishing, 2023.
[5] Ling,Q.,&Isa,N.a.M.(2023). \"PrintedCircuitBoardDefectDetectionMethodsBasedonImageProcessing,MachineLearningandDeepLearning:A Survey.\"IEEE Access, 11, 15921–15944. DOI: 10.1109/access.2023.3245093.
[6] Singh,S.,Ramya,R.,Sushma,V.,Roshini,S.,&Pavithra,R.(2019).\"FacialRecognitionUsingMachineLearningAlgorithmsonRaspberryPi.\"20215thInternationalConferenceonElectrical,Electronics,Communication,ComputerTechnologiesandOptimizationTechniques(ICEECCOT),197–202.DOI: 10.1109/iceeccot46775.2019.9114716.
[7] Chaudhari A., Upganlawar V., Barve T., Vaidya R. and Shelke D., \"Analysis of YOLO V3 for Multiple Defects Detection in PCB,\" 2024 ParulInternational Conference on Engineering and Technology (PICET), Vadodara, India, 2024, pp. 1-6, doi: 10.1109/PICET60765.2024.10716153.
[8] Chaudhari,A.,Manwatkar,V.,Shinde,G.(2025).AnalysisofYOLOv5forDetectionofMultipleDefectsinPCB.In:Kumar,A.,Pachauri,R.K.,Mishra,R.,Kuchhal,P.(eds)IntelligentCommunication,ControlandDevices.ICICCD2024.LectureNotes inNetworksandSystems, vol1164.Springer,
[9] Singapore.https://doi.org/10.1007/978-981-97-8329-8_35
[10] Marot, J., & Bourennane, S. (2017). \"Raspberry Pi for Image Processing Education.\"2021 29th European Signal Processing Conference (EUSIPCO).DOI: 10.23919/eusipco.2017.8081633.
[11] S, N. M., P, N. N., P, N. I., P, N. M. S. T., & R, N. M. (2021). \"Object and Lane Detection for Autonomous Vehicle Using YOLO V3 Algorithm.\"AIPConference Proceedings, 2387, 140009. DOI: 10.1063/5.0068836.
[12] Chhetri, S. P., Bhat, S., Timalsina, P., & Magar, B. T. (2023). \"Detection of Missing Component in PCB Using YOLO.\"International Journal onEngineering Technology, 1(1), 62–71. DOI: 10.3126/injet.v1i1.60902.
[13] Wu, X., Ge, Y., Zhang, Q., & Zhang, D. (2021). \"PCB Defect Detection Using Deep Learning Methods.\"2022 IEEE 25th International ConferenceonComputer Supported Cooperative Work in Design (CSCWD), 873–876. DOI: 10.1109/cscwd49262.2021.9437846.
[14] Li, Q., Zheng, Q., Jiang, S., Hu, N., & Liu, Z. (2024). \"An Improved YOLOv5-Based Model for Automatic PCB Defect Detection.\"Journal of PhysicsConference Series, 2708(1), 012017. DOI: 10.1088/1742-6596/2708/1/012017.
[15] L. Raju, G. Sowmya, S. Srividhya, S. Surabhi, M.K. Retika, and M. Reshmika Janani, \"Advanced Home Automation Using Raspberry Pi and MachineLearning,\"2021 7th International Conference on Electrical Energy Systems (ICEES), Chennai, India, 2021, pp. 600-605. DOI:10.1109/ICEES51510.2021.9383738.