Vision-based Assembly and Inspection System for Golf Club Heads

Highlights

• This study presents a Vision-based Assembly and Inspection System for Golf Club Heads (VAIS-GCH).

• Ths system adopts automated mechanical arms matched with cameras for 3D space alignment.

• Through the coordination of two cameras, the accuracy of the golf club head assembly is increased significantly.

Abstract

Golf club head production is started by creating metallic films and forming blank casts, which then undergo complex and sophisticated assembly procedures, morphological inspection and strength testing. Not only chemicals used in these processes often threaten the health of operators, but also most of this backend processing involves manual assembly and welding, demanding substantial amounts of manpower. In light of these pitfalls, this paper proposes a Vision-based Assembly and Inspection System for Golf Club Heads (VAIS-GCH) to mitigate the time-consuming and labor-intensive golf club head production process. Two cameras are coordinated to capture top-view image of the striking plate, and top- and side-views of the casting body for visual processing. After the barycenter shift and shifting angle of the striking plate are determined, a robot arm is then directed to suction up the striking plate from the conveyor belt, adjust it to the canonical orientation, and move the yet-to-be assembled striking plate to the top of the casting body. The 3D spatial positions of the casting body, namely, the XYZ shifting angles of the coupling opening, are detected by comparing the top- and side-view images captured with pre-stored 3D golf club head template to facilitate the coupling of the striking plate and casting body. The side-view camera monitors the insertion depth of the striking plate during the coupling process to make sure the surface of the striking plate levels with the casting body. The loft angle, formed by the line perpendicular to the tangent of the bottom of casting body and the tangent of the striking surface, is tested to confirm whether specifications are satisfied. Through the coordination of two cameras, the accuracy and efficiency of the golf club head assembly are increased significantly. The alignment error is smaller than 0.8 mm and the assembly and inspection process takes less than 2 s.

Introduction

Products in traditional 3D (Dangerous, Dirty and Difficult) industries typically consist of assembling multiple metal parts through screws, welding, and adhesives. A substantial amount of manual work and human-operated machinery are often indispensable in most production processes. Demands on the technical level of manpower are high; yet yields are only 60–80%, leading to high costs and unstable quality. This is especially true for the manufacturing of golf clubs. Targeting on various player age groups, the models of golf clubs are diverse in terms of flying direction and distance after ball striking. Yet the quantity manufactured for each specification tends to be small with relatively high unit profit and short product life cycle. Traditionally, manufacture of golf club heads are completed manually by performing fixture production, welding, and testing steps to ensure production quality. Following preliminary assembly, loft angles of golf club heads are examined by applying rulers. Human visual inspection is then conducted to verify whether products conform to customer specifications. This is a laborious process with low efficiency, and human errors are common. Golf club manufacturers routinely adjust the assembling procedures and inspection criteria in response to different specifications demanded by various models within a rather short product cycle. This practice inevitably leads to increasing human operating errors. Take the fabrication of a high-end titanium golf club head for example, the head is composed of a casting body, manufactured by die-casting, and a striking plate, finished by forge and EDM wire cut. The error regarding the outline of the striking plate fabricated is negligibly small due to the high precision of the wire cut machine employed. However, the casting body may be warped due to anisotropic shrinkage during cooling down. The distortion of casting body makes the pre-taught welding path no longer fit for the real shape of weld seam, resulting a drop of 8–10% in yielding rate. In light of the high cost of the titanium, the integration of robotic arm with vision means becomes imperative to circumvent this problem.

Robotic arms are often employed for automated processes in industrial applications. For example, spot welding is first performed by robots in automobile industry because the simple joint space interpolation involved. For seam weld, a smooth curve in Cartesian space should be generated from prescribed positions. This is fulfilled by a laborious teaching and complex coordinate transformations. An adequate robotic moving path should include velocity and acceleration information important to tasks such as seam welding, spray painting and assembly, etc. This requirement cannot be fulfilled simply by manual teaching. Offline programming through a CAD system is often resorted to alleviating the path planning problems [1], [2], [3], [4]. When the shape of a coupling workpiece is distorted, the robot should be able to dynamically adapt to the new outline of the weld seam. Recent development of machine vision technology is employed to inspect the welding surface, and to adjust the robot path during welding [5], [6], [7]. Features are identified after image capture and utilized to calculate corresponding 3D space coordinates and angles of rotation for fast, dynamic planning of smooth paths. This allows mechanical arms to reach the correct position accurately.

Robotic arms combined with machine vision were employed to construct an automatic golf club head welding system [8]. The image of golf club head that had undergone preliminary welding was processed by Sobel and Laplace filters for boundary separation. Mobile paths and shifting angles were then computed to guide the robotic arms to perform automatic welding operation. In this setup, the critical stages before and after preliminary welding, e.g., coupling of striking plate and casting body, and loft angle inspection, still demand human operation, nullifying a fully automatic golf club head manufacturing process. Although robotic arms were employed to expedite 3D object׳s grasp and movement, these applications did not demand space coupling and component coupling in high precision. The aforementioned technological transition is further hindered by the 3D space curved surface construction encountered in most metal parts, which are far more complex than simple planar objects. This causes difficulty in performing high-precision automated coupling, welding, and assembly. Therefore, the vision-based 3D correspondence and coupling for spatial objects has become a key issue in developing intelligent automatic equipment [9], [10], [11], [12], [13]. Also, a nondestructive inspection approach is required for the mass production of components for industrial applications, such as the automobile, shipbuilding, and aerospace industries where high inspection rate is required [14], [15], [16], [17], [18], [19].

This paper presents a Vision-based Assembly and Inspection System for Golf Club Heads (VAIS-GCH) to resolve these deficiencies. This system is equipped with two cameras, one mounted on a robot arm to capture top-view images of the striking plate and casting body, and the other a stationary camera positioned laterally to the casting body to provide the corresponding side view. Images captured at different stages of assembly are further processed for 3D alignment between the striking plate and casting body. The camera on the moving robotic arm captures the striking plate image, conveyed through an assembly line, from the top to derive the barycenter position and shifting angle of the striking plate. The robotic arm then suctions up the striking plate from the barycenter and compensates the angle shifted to a preset canonical orientation. After moving the robotic arm with the striking plate to the top of the casting body, both the camera on the robotic arm and the stationary one take a top-view and side-view image of the casting body, respectively. The top-view image of the casting body is utilized to calculate the barycenter of the casting body. The pre-stored 3D golf club head template is then compared to derive the XYZ-plane shifting angles of the casting body within a fixture structure. The robotic arm then inserts the striking plate into the opening of casting body based on the 3D spatial position derived. During the coupling process, the stationary camera monitors the insertion depth to make certain the top surface of the striking plate levels with that of the casting body. The side view after coupling of striking plate and casting body is then analyzed to determine whether the loft angle, formed by the line perpendicular to the tangent of the bottom of casting body and the tangent of the striking surface, meets the specification. This can greatly improve coupling precision between the striking plate and casting body, thereby increasing production and inspection efficiency to expedite fast transitions between different models and assembly lines.