Last Updated: May 7, 2026

Crankshaft Robotic Bin Picking with 3D Vision: Engine Part Sorting and Feeding Video

Crankshaft robotic bin picking automates one of the harder handling steps in engine manufacturing: selecting heavy, irregular crankshafts from a deep bin where every part may be stacked at a different angle. The video below shows how a 6-axis robot, 3D vision, pose estimation, and a flexible gripper can turn random crankshaft sorting into a repeatable feeding process for downstream machining, marking, or assembly stations.

If the embedded player does not load, open the video directly on YouTube: Crankshaft Robotic Bin Picking with 3D Vision.

Process Overview

The crankshaft is the core power transmission component of an engine. In automotive manufacturing, it moves through forging, precision turning, grinding, dynamic balancing, marking, and assembly-related processes. Between these stations, sorting and feeding must be reliable enough to support stable takt time without wrong-part feeding or repeated operator intervention.

Traditional manual sorting is slow and error-prone because crankshafts have irregular geometry, offset counterweights, reflective machined surfaces, and random stacked poses. Operators must judge both the part model and the best grasp point by eye, which makes cycle time and quality stability difficult to control.

The current engineering solution uses 6-axis robots paired with 3D vision systems. Laser structured-light scanning generates dense point clouds, while deep learning supports part recognition and pose estimation for random bin picking. A well-tuned cell can reach about 48 seconds per piece, and a single flexible gripper can handle 6 to 8 crankshaft models without hardware changeover. For similar material handling use cases, see EVSINT’s handling robot category.

Common Pitfalls in Crankshaft Bin Picking

Inadequate camera field of view and depth of field. Bin depths often reach 800 mm. If a project uses a narrow-FOV camera to reduce cost, the bottom layer may not be fully covered, which keeps recognition failure rates high and causes frequent line alarms.

Poor point cloud quality on reflective surfaces. Precision-machined crankshaft surfaces can create specular reflection during laser scanning. The result is point cloud voids and noise, directly reducing recognition accuracy and pose estimation reliability.

Insufficient deep learning training data. New crankshaft models may enter production without enough sample images. Weak model generalization can extend field commissioning from an expected 3 days to 2 weeks.

Trajectory planning without collision detection. Some systems rely on basic kinematic simulation and do not fully consider bin walls, neighboring parts, and gripper interference. In deep-bin picking, this can create collision-triggered downtime and repeated manual resets.

No secondary orientation station. Direct placement into a downstream fixture may fail to meet pose accuracy requirements. A secondary station can correct orientation before machining, marking, or conveyor transfer. This is especially important when the cell feeds a machine tending robot process or CNC-related station.

Core Technical Solutions

3D vision scanning. Laser structured-light or laser line scanning is recommended. The field of view should cover the entire bin, and depth of field should be at least 800 mm. Active laser projection helps suppress ambient light interference, while hardware-level HDR improves point cloud quality on reflective surfaces.

Deep learning recognition. Attention-based point cloud segmentation can identify the topmost pickable target even under severe occlusion and random stacking. Few-shot transfer learning helps shorten new-model commissioning to 1 to 2 days when enough representative samples are collected.

Intelligent trajectory planning. The cell should build a 3D obstacle scene from real-time point cloud geometry, then combine it with robot kinematics to generate collision-free pick and place paths. An eye-in-hand camera architecture allows the robot to adjust the scanning angle and reduce blind spots.

Flexible gripper design. A pneumatic and electric hybrid gripper can support the crankshaft balance weight while lateral positioning pins adapt to journal-spacing changes. The gripper is the core mechanical interface, so it should be designed together with the robot end effector and downstream fixture rather than treated as an accessory.

Standardized Engineering Layout

A crankshaft robotic bin picking cell usually integrates a 6-axis robot, flexible gripper, structured-light 3D vision, secondary orientation station, laser marker, safety system, and electrical control cabinet. The workflow is straightforward: the bin enters the station, the robot scans with an end-mounted camera, deep learning identifies the model and pose, trajectory planning guides the pick, the secondary station corrects orientation, the laser engraves a 2D code if required, and the part is placed on the conveyor or fixture.

For sorting and transfer processes where parts are not as heavy as crankshafts, a pick and place robot may be enough. For crankshafts, the payload, reach, wrist torque, and gripper center of gravity usually make a 6-axis industrial robot the safer baseline.

The safety system should combine perimeter fencing, interlocked gates, light curtains, and safety PLC logic. The electrical cabinet should use industrial cooling so internal temperature remains below 40 degrees C, reducing the risk of VFD derating and unstable robot or conveyor response.

Technical Summary

Crankshaft sorting automation depends on three linked capabilities: vision stability, collision-free trajectory planning, and gripper flexibility. Vision stability determines the recognition ceiling. Trajectory planning determines the collision risk ceiling. Gripper flexibility determines how efficiently the cell can change between crankshaft models.

The hardware layer is already mature. The real bottleneck is field adaptation: deep learning model tuning, reflective-surface handling, and rapid migration across multiple crankshaft models. When selecting an integrator, automotive project experience matters more than comparing only camera resolution or robot brand. After deployment, plants should iterate vision models and trajectory libraries every 3 to 6 months to push OEE above 90%.

Last Updated: May 7, 2026