DaoAI 3D Robot Vision for Automotive Fastener Tightening Defect Detection

WeLinkirt DaoAI 3D Robot Vision (proprietary 3D camera + 6D pose estimation, bin picking, glue dispensing/assembly/loading/unloading guidance, brain-eye-body closed-loop, sub-millimeter hand-eye coordination) accurately detects the proper tightening and anti-loosening marks of critical fasteners in automotive powertrain systems, reducing the missed detection rate from 1.5% to <0.2%, significantly improving product quality and production line efficiency.

WeLinkirt DaoAI 3D Robot Vision (proprietary 3D camera + 6D pose estimation, bin picking, glue dispensing/assembly/loading/unloading guidance, brain-eye-body closed-loop, sub-millimeter hand-eye coordination) accurately detects the proper tightening and anti-loosening marks of critical fasteners in automotive powertrain systems, reducing the missed detection rate from 1.5% to <0.2%, significantly improving product quality and production line efficiency. In the automotive and parts manufacturing sector, the correct installation of fasteners is paramount to ensuring vehicle performance and safety. Particularly in the assembly of core powertrain components like engines, transmissions, and chassis, the tightening torque, seating depth, and integrity of anti-loosening marks for bolts and nuts directly impact vehicle reliability. With the rapid development of new energy vehicles, there's an increasing demand for lightweight, integrated, and highly reliable components. Traditional manual inspection or indirect detection methods based on torque sensors are no longer sufficient to meet the increasingly stringent quality control standards and production cycle times. For instance, on the production line of a leading automotive powertrain supplier, the assembly of a new hybrid transmission housing requires comprehensive inspection of dozens of critical fasteners to ensure each bolt is precisely seated and has the correct anti-loosening adhesive, preventing loosening due to vibration or thermal expansion/contraction, which could lead to serious safety hazards.

Pain Points: Why This Hurdle Was So Difficult

The client faced multiple challenges in the fastener tightening inspection process. Firstly, manual visual inspection was inefficient and inconsistent, prone to human fatigue, leading to a missed detection rate as high as 1.5% and a false positive rate of around 8%, with significant time spent on manual re-inspection. Secondly, traditional 2D vision solutions struggled to accurately identify the height and depth of fasteners, or the 3D form of anti-loosening marks, in complex curved surfaces and obstructed environments. They were highly sensitive to lighting and product surface reflections, often failing due to environmental changes. Thirdly, the wide variety of product models, fastener types (including different specifications of bolts, nuts, and rivets), and their placement in confined spaces, sometimes with overlapping multi-layer structures, exponentially increased inspection difficulty. Each model changeover required several hours to readjust camera positions and parameters, severely impacting production line flexibility.

From a process perspective, the proper seating of a fastener involves not just its planar position but also axial depth information, such as whether a bolt is fully threaded in or if it's floating. Anti-loosening marks (e.g., adhesive, paint dots) also need to be checked for complete coverage over the thread or bolt head, without overflow or absence. These subtle 3D morphological differences often appear as blurry edges or light/shadow variations in 2D images, making them prone to misjudgment. From an imaging perspective, metal surfaces are highly reflective, and fasteners themselves are often cylindrical, leading to specular reflections that can cause overexposure or underexposure, losing critical details. Additionally, when fasteners are located in deep holes or recesses, traditional light sources struggle to illuminate them effectively, creating shadows that further complicate detection. From a cycle time perspective, automotive production lines demand extremely high efficiency, with single-piece inspection times typically in seconds. Traditional scanning methods for complex 3D inspection are too time-consuming to meet these cycle requirements. The current industry hot topic focuses on the “ChatGPT moment” for embodied AI, as demonstrated by the Ecovacs disassembly robot. Intelligent robots need precise “brain-eye-body” closed-loop capabilities to perform accurate operations and perception in the complex physical world. Fastener inspection is precisely an ultimate test of this “embodied intelligence,” requiring robot vision systems not only to “understand” the 3D world but also to guide robotic arms for high-precision judgment and operation, which is where traditional solutions fall short.

Technical Principles

The WeLinkirt DaoAI 3D Robot Vision system fundamentally addresses these challenges through its proprietary high-precision 3D camera combined with advanced 6D pose estimation algorithms. Our 3D camera utilizes structured light projection technology, emitting specific pattern gratings and capturing surface deformation from multiple industrial cameras at different angles to reconstruct high-precision point cloud data. Compared to traditional laser line scanning or time-of-flight (ToF) technologies, structured light acquires 3D information faster and more densely, especially suitable for detecting complex geometries and minute features. The system processes point cloud data using deep learning algorithms, achieving sub-millimeter accuracy. The 6D pose estimation module, based on deep learning models, directly extracts the X, Y, Z translational dimensions and Roll, Pitch, Yaw rotational dimensions of the target fastener from the 3D point cloud data, precisely determining its spatial position and orientation, even in disorganized or randomly placed scenarios.

Compared to traditional rule-based 2D AOI, our 3D robot vision offers significant advantages. 2D AOI relies on 2D features like image grayscale, color, and edges, making it susceptible to variations in lighting, reflections, occlusions, and product surface textures, with limited ability to identify 3D morphological defects. In contrast, DaoAI 3D Vision directly acquires complete 3D geometric information of the object, enabling precise measurement of key dimensions such as fastener height, depth, and flatness, directly determining if a bolt is seated correctly or if anti-loosening adhesive is fully covered. For example, for a floating bolt issue, 2D vision might misinterpret due to light/shadow changes, but 3D vision can directly measure the distance between the bolt head's top surface and a reference plane, providing precise millimeter-level height data. Furthermore, combined with “brain-eye-body” closed-loop control, the system provides real-time visual data feedback to the robotic arm, guiding it to precise inspection positions, achieving sub-millimeter hand-eye coordination, ensuring detection stability and repeatability. Compared to manual visual inspection, the system eliminates human subjective judgment errors and fatigue, elevating detection consistency to industrial standards, while increasing detection speed severalfold, significantly reducing labor costs and production risks.

Typical Application Scenarios

• **Bolt/Nut Tightening Seating Detection:** For critical components like engine blocks and transmission housings, inspecting whether bolts/nuts are fully tightened, or if there's floating or insufficient threading. The challenge lies in the tight fit tolerance between the bolt head and mounting hole, often located in deep holes or multi-layer structures, requiring high-precision 3D measurement to exclude planar occlusion interference.
• **Anti-Loosening Mark (Adhesive/Paint Dot) Integrity and Position Detection:** Checking if anti-loosening adhesive or paint dots on bolt heads or threads are applied completely according to process requirements, without missing, overflowing, and in the correct position. The difficulty lies in the irregular thickness, color, and shape of the adhesive, as well as metal surface reflections interfering with 2D images. 3D vision can accurately reconstruct the 3D morphology of the adhesive to determine its coverage and volume.
• **Rivet Installation Quality Inspection:** In scenarios like body connections and interior component fastening, inspecting if rivets are fully set, without looseness, deformation, breakage, or uneven protrusion. The challenge is the diverse shapes of rivet heads, and they may be flush with the substrate surface after riveting, which is difficult for 2D to distinguish. 3D vision can measure minute deformations and height differences of rivet heads.
• **Foreign Object Detection and Missing Part Detection:** After assembly, scanning the area around fasteners to check for left-behind tools, impurities, or missing washers, clips, etc. The difficulty lies in irregular shapes of foreign objects, usually small in size. 3D vision can identify abnormal morphologies beyond the standard model through point cloud analysis.

Case Study

A leading Tier-1 automotive component supplier, specializing in the R&D and manufacturing of high-end powertrain systems, had a newly built hybrid transmission assembly line with extremely high demands for product quality and production efficiency. Before integrating the WeLinkirt DaoAI 3D Robot Vision system, the line primarily relied on manual spot checks and some 2D vision assistance for fastener tightening inspection on transmission housings. Due to the complex internal structure of the transmission, with dozens of fasteners, some located in hard-to-see deep areas, the missed detection rate reached 1.5%, leading to millions of RMB in rework and quality claim costs monthly due to fastener issues. Simultaneously, the 2D vision system frequently generated false positives when identifying the integrity of anti-loosening adhesive due to reflections or adhesive color variations, requiring hundreds of manual re-inspections daily and consuming significant labor hours. To enhance the precision and automation of inspection, the client adopted our DaoAI 3D Robot Vision solution.

During the implementation, we first conducted 3D data acquisition and model training for dozens of different specifications of fasteners and anti-loosening adhesive types. Based on the unified foundation of the WeLinkirt DaoAI World Model, we used a small number of good samples (1-20 good samples) for training, quickly building robust detection models. The system was deployed at the end of a robotic arm, which then executed programmed path planning to scan and inspect all critical fasteners on the transmission housing one by one. After deployment, the system achieved 100% full inspection, with a detection rate of 99.4% for defects such as floating bolts, insufficient threading, and missing or overflowing anti-loosening adhesive, reducing the missed detection rate to <0.2%. The false positive rate also significantly decreased from the original 8% to 2.9%, greatly reducing the workload for manual re-inspection. Furthermore, because the 3D vision system is insensitive to lighting and product surface reflections, model changeovers only required selecting the corresponding product model in the software, without needing to readjust physical camera positions or light sources. Changeover time was reduced from several hours to 5min, significantly improving production line flexibility.

The WeLinkirt DaoAI 3D Robot Vision system, with sub-millimeter precision and 'brain-eye-body' closed-loop intelligence, brings unprecedented reliability and efficiency improvements to automotive component assembly, truly transforming the potential of embodied AI into productivity.

WeLinkirt Solution and Products

The WeLinkirt DaoAI 3D Robot Vision system is the core of this solution. Our proprietary high-precision 3D camera rapidly acquires high-density point cloud data of object surfaces, combined with powerful 6D pose estimation algorithms, enabling precise identification and positioning of complex workpieces. In practical deployment, we offer various integration methods such as SDK/API/Docker, supporting 100% local private deployment to ensure customer data security remains on-site. For new products or defect types, customers can leverage the APDT positive/few-shot learning capabilities of the DaoAI AI AOI software system, requiring only 1-20 good sample images to complete model training and automatic programming within 5 minutes, significantly shortening changeover and debugging times. Additionally, integrated with the DaoAI World Model's unified foundation, the system possesses semantic understanding and cross-scenario generalization capabilities, continuously learning and optimizing from production line feedback to improve detection accuracy and efficiency. Throughout this process, the robot vision system acts as the “eyes” and “brain,” guiding the robotic arm's “body” to perform precise inspection tasks, forming a complete embodied AI closed-loop.

In this application, the system reduced the missed detection rate of fastener tightening defects from 1.5% to <0.2%, and the false positive rate by -63%, saving the client millions of RMB annually in quality rework and claim costs. Concurrently, changeover time was dramatically reduced to 5min, allowing the production line to respond more flexibly to multi-variety, small-batch production demands, enhancing overall production efficiency and market responsiveness. High-precision, high-efficiency, and highly flexible automated inspection not only significantly improved product quality stability but also helped the client maintain a leading position in the fiercely competitive automotive component market, effectively mitigating compliance risks and brand reputation losses due to critical component failures.