The relentless drive towards miniaturization and increased precision in modern manufacturing, spanning sectors from aerospace to automotive, has created a significant technological bottleneck: the automated assembly of fasteners within extremely confined and geometrically complex internal spaces. Traditional robotic end effector solutions for bolt tightening are often too bulky, heavy, or inflexible to operate effectively in such environments. Manual intervention in these scenarios is not only ergonomically challenging and labor-intensive but also prone to inconsistencies in achieving precise torque specifications, ultimately compromising product quality and reliability. While existing research has explored specialized end effectors for side-access tightening or utilized advanced learning algorithms for robotic disassembly, these approaches often involve complex mechanisms, significant weight, or reliance on external sensing that limits their practical application in dense assemblies. There is a pressing need for a fundamentally new approach—a lightweight, compact, and intelligent robotic end effector specifically engineered for the automated installation, tightening, and removal of bolts in狭小空间. This work presents the design and implementation of such an end effector, focusing on a novel mechanical architecture that enables self-contained bolt handling and a sophisticated current-based dual-loop control system that ensures precise torque delivery without the need for bulky external sensors.
The core philosophy behind this end effector is the integration of bolt handling functions directly into a minimized form factor. The mechanical design prioritizes a cylindrical profile with minimal diameter to facilitate access. The end effector’s functionality is decomposed into three integrated subsystems: the drive unit and housing, the power transmission and automated engagement mechanism, and the interchangeable magnetic socket. The overall axial length of the mechanical body is 58 mm, with a maximum diameter of 24 mm, resulting in a total mass of approximately 150 grams when integrated with its drive unit. This compactness is a direct enabler for deployment in confined spaces.
The bolt handling subsystem is centered on an interchangeable magnetic socket. For targeting standard M4, M5, and M6 external hexagonal carbon steel bolts, a magnetic coupling provides a simple, reliable, and lightweight grasping solution. Commercially available magnetic hex socket bits are modified to serve as the primary interface. One end is machined with external threads for secure and quick attachment to the end effector’s drive shaft, while a transverse setscrew hole ensures positive rotational locking. This design allows for rapid tool changeover to accommodate different bolt sizes, a critical feature for flexible automation. The magnetic force, approximately 1.5 N, is sufficient to retain the bolt during positioning but is overcome during the controlled tightening process.
The power transmission system must reliably transfer torque from the rotary actuator to the socket while integrating the mechanism for automatic axial engagement. A concentric sleeve design is employed. An inner drive shaft is directly coupled to the actuator. This shaft is connected to the magnetic socket. An outer sleeve assembly, split into upper and lower guide halves for ease of assembly, houses the automatic engagement mechanism. The key innovation lies in the integration of a helical groove and tension spring system for autonomous axial motion. A close-wound tension spring is wound into matching helical grooves on the inner drive shaft and a fixed post within the outer sleeve. During the bolt installation phase, the robotic arm positions the end effector such that the magnetic socket picks up a bolt. As the arm then moves the end effector towards the target hole, the socket and inner shaft can retract axially against the spring force, allowing the bolt to align and start threading without requiring precise independent axial motion from the robot arm itself. Conversely, after tightening or during removal, the spring automatically returns the socket to its extended position, ejecting or disengaging the bolt. The spring parameters are critical. The stiffness P is given by:
$$ P = \frac{G d^4}{8 D^3 n} $$
where G is the shear modulus (71 GPa for 304 stainless steel), d is the wire diameter, D is the mean coil diameter, and n is the number of active coils. The required spring extension l is greater than the threaded length of the longest target bolt (12 mm) but less than the travel limit (16 mm). Solving for a design that provides the necessary force F (1.5 N) at the required extension yields a spring with d = 0.5 mm, D2 (outer diameter) = 8 mm, and a free length of 10.5 mm. This mechanism significantly simplifies the trajectory planning and control for the host robotic manipulator, as it only needs to position the end effector within a coarse envelope, while the fine axial alignment is handled passively by the end effector itself.
To validate the structural integrity of the end effector under maximum load, stress simulations were performed for both the tightening and disassembly operations. A driving torque of 7 N·m was applied, representing the peak output of the selected actuator. The results confirmed that the maximum von Mises stress in all components, primarily constructed from aluminum alloy, remained well below the material’s yield strength. Furthermore, deformation was negligible, ensuring precise alignment and reliable torque transmission. The successful simulation underpins the robustness of the lightweight design.
| Parameter | Value | Description |
|---|---|---|
| Axial Length | 58 mm | Body length excluding actuator |
| Max Diameter | 24 mm | Maximum outer body diameter |
| Total Mass | ~150 g | Including drive unit and housing |
| Bolt Sizes | M4, M5, M6 | Target external hex carbon steel |
| Spring Stiffness | ~0.125 N/mm | Calculated tension spring rate |
| Magnetic Force | ~1.5 N | Approximate bolt retention force |
The control system for this specialized end effector is paramount for achieving precise torque control, which is the ultimate measure of a successful bolted joint. Instead of employing a direct torque sensor—which would add weight, complexity, and bulk—the system utilizes an indirect method based on motor current feedback. For a permanent magnet DC motor (or a brushless DC motor driven in a similar fashion), the electromagnetic torque T is directly proportional to the armature current I:
$$ T = C I $$
where C is the motor’s torque constant. By precisely controlling and regulating the current supplied to the motor, one can effectively control the output torque at the end effector’s socket. This principle forms the core of the control strategy.
The hardware architecture is built around an STM32F103RCT6 microcontroller unit (MCU), chosen for its computational capability, multiple analog and digital interfaces, and real-time performance. The actuator is a high-torque 360° continuous rotation digital servo (70 kg·cm), which integrates a DC motor, gearbox, and internal drive circuitry. The control system employs a dual closed-loop configuration for enhanced stability and accuracy. The inner loop is a fast current control loop. A current sensing board based on the MAX472 chip measures the real-time current draw of the servo motor, converts it to a proportional voltage signal, and feeds it back to an ADC pin on the STM32 MCU. The outer loop is a voltage control loop. A voltage sensor monitors the actual voltage supplied to the servo motor-driver system. This signal is also fed back to the MCU. The MCU generates the primary control signal via a Digital-to-Analog Converter (DAC) output, which is fed into a MOSFET-based opto-coupled voltage control module. This module regulates the power supplied to the servo system based on the DAC command. Four independent buttons are connected to external interrupt (EXTI) pins on the MCU, allowing the operator to select operating modes (e.g., tighten/loosen) and target torque/current setpoints for different bolt sizes.
The control law for both loops utilizes a discrete-time Proportional-Integral-Derivative (PID) algorithm implemented in the MCU’s firmware. The general form of the PID controller output u(t) is:
$$ u(t) = k_P e(t) + k_I \int_0^t e(t) dt + k_D \frac{de(t)}{dt} $$
where e(t) is the error (setpoint minus feedback) at time t, and kP, kI, kD are the proportional, integral, and derivative gains, respectively. In this cascaded structure, the outer voltage PID controller ensures a stable supply voltage to the motor drive system, compensating for line variations and load changes. The inner current PID controller rapidly adjusts the command to track the desired current setpoint, which is directly correlated to the target torque. This two-degree-of-freedom approach provides superior disturbance rejection compared to a single-loop system.
| Component | Model/Type | Primary Function |
|---|---|---|
| Microcontroller | STM32F103RCT6 | Main processor for signal processing and control logic |
| Actuator | 360° Continuous Rotation Servo (70 kg·cm) | Provides rotary torque for tightening/loosening |
| Current Sensor | YYl-2 Module (MAX472 based) | Measures real-time motor current for inner control loop |
| Voltage Sensor | Risym Voltage Detection Module | Measures supply voltage for outer control loop |
| Power Control | MOSFET Opto-coupled Module | Acts as a voltage-controlled power switch based on DAC input |
| Interface | 4x Independent Push Buttons | Provides user input for mode and setpoint selection |
Prior to physical implementation, the proposed control strategy was rigorously modeled and simulated in MATLAB/Simulink. The simulation model incorporated the dynamic model of the servo motor, the PID controllers for both voltage and current loops, a model of the bolt tightening process (where load torque increases with angle/pretension), and the sensors. The objective was to tune the PID parameters to achieve a fast, stable response with minimal overshoot and steady-state error. For a target tightening torque of 4.470 N·m, corresponding to a calculated current setpoint of 3.51 A, the simulation results were promising. The electromagnetic torque from the motor closely tracked the rising load torque during the tightening phase. The final steady-state electromagnetic torque settled at 4.461 N·m, demonstrating an error of less than 1% from the target. Furthermore, the motor torque stabilized slightly before the load torque, indicating a responsive and well-tuned control system. This simulation validated the feasibility of the current-based torque control approach and provided initial PID gains for the physical system.
The firmware for the STM32 MCU was developed using a combination of STM32CubeMX for hardware initialization and Keil MDK for code generation and compilation. The control logic was partially modeled in Simulink and automatically generated into efficient C code using the Embedded Coder, ensuring a seamless transition from simulation to real-time execution. The software handles ADC sampling for current and voltage feedback, implements the dual discrete PID controllers, manages the DAC output to the power module, processes button interrupts to change operational states, and controls the servo’s direction via a PWM signal. Key tunable parameters, such as PID gains, filter coefficients, and ADC sampling rates, were made accessible for final calibration on the hardware.
The performance of the complete robotic end effector system was evaluated on a dedicated test platform. The end effector was mounted on a collaborative robot arm. A rigid test fixture held a metal plate with threaded holes. A precision annular load cell, installed between the bolt head and the plate, measured the actual clamping force (preload) during tightening. For each bolt size (M4, M5, M6), a target preload was calculated from standard torque-preload relationships assuming a constant friction coefficient. The end effector, commanded via its button interface for the specific bolt size and tighten operation, performed 20 consecutive tightening cycles for statistical analysis.
The results demonstrated the system’s capability for precise and repeatable torque delivery. For M4 bolts (target preload 2535 N), the average achieved preload was 2504.8 N with a standard deviation of 15.1 N. For M5 bolts (target 4085 N), the average was 4047.7 N (std dev 18.8 N). For M6 bolts (target 5720 N), the average was 5684.3 N (std dev 17.6 N). In all cases, the coefficient of variation (discreteness) was below 0.01, indicating high repeatability. The consistent slight negative bias between the average achieved preload and the target can be attributed to factors such as the inertia of the end effector’s rotating parts, minor misalignments in the test setup, and potential variations in the thread friction, which are not accounted for in the simple current-to-torque model. Nevertheless, the repeatability is the more critical metric for automated assembly, as a consistent bias can be compensated by calibrating the current setpoint.
| Bolt Size | Target Preload (N) | Average Achieved Preload (N) | Standard Deviation (N) | Coefficient of Variation |
|---|---|---|---|---|
| M4 | 2535 | 2504.8 | 15.1 | 0.0060 |
| M5 | 4085 | 4047.7 | 18.8 | 0.0047 |
| M6 | 5720 | 5684.3 | 17.6 | 0.0031 |
A deeper analysis of the tightening process data confirms the underlying control principle. Plotting the motor current against the measured preload from a representative tightening cycle for each bolt size reveals a strong linear correlation once the bolt is engaged and the preload begins to rise. This linear relationship validates the fundamental equation $$ T = C I $$, as the load torque (which is a function of preload) is proportional to the motor current. Furthermore, as expected, for a given preload value, the required current increases with bolt size (from M4 to M6), and the slope of the current-preload line also increases, reflecting the larger torque required to achieve the same preload in a larger, stiffer bolt. This data empirically verifies the effectiveness of the current-controlled end effector for managing different fastener specifications.
In conclusion, this work has successfully addressed the challenge of automated bolt assembly in confined spaces through the design and implementation of a novel, compact robotic end effector. The mechanical design integrates an automatic axial engagement and retraction mechanism via a helical spring system, dramatically simplifying the path planning requirements for the host robot. The control system innovates by employing a voltage-current dual closed-loop strategy, using motor current as a precise proxy for output torque, thereby eliminating the need for a bulky and expensive torque sensor embedded in the end effector. Experimental validation on M4, M5, and M6 bolts confirmed the system’s ability to perform installation, tightening to a specified torque, and removal with high repeatability. The developed end effector represents a significant step towards true automation in restrictive environments, such as inside engine compartments, airframes, or complex machinery. Future work will focus on enhancing the control algorithm to incorporate angle monitoring for a combined torque-angle tightening strategy, improving the accuracy of the preload prediction model by accounting for friction variations, and further miniaturizing the design to tackle even smaller fasteners and more restrictive spaces.