Robotic Grinding Achieves Precision with PI Control Innovation
In the evolving landscape of industrial automation, achieving consistent material removal during robotic surface finishing has long posed a significant challenge—particularly when dealing with high-hardness cast steel and iron components. Traditional robotic grinding systems often falter due to inherent inaccuracies in workpiece dimensions and pre-programmed trajectories, leading to uneven surface finishes, tool overloads, or incomplete material removal. Now, a breakthrough control strategy developed by researchers at Qingdao University is transforming how industrial robots maintain uniform grinding depth in real time—without relying on fragile force sensors or complex adaptive algorithms.
Led by Du Longfei and Wang Yu from the School of Mechanical and Electrical Engineering at Qingdao University, the research introduces a novel PI (Proportional-Integral) control-based method that dynamically adjusts the robot’s Tool Center Point (TCP) to ensure consistent grinding thickness. Published in China Mechanical Engineering under the title “Research on iso-thickness grinding by robot based on PI control,” the study presents a robust, field-tested solution tailored specifically for the harsh conditions of metal casting workshops.
Unlike conventional approaches that depend on six-axis force sensors to detect contact forces, this new method leverages the torque output of a servo-driven grinding motor as a proxy for grinding load. The rationale is straightforward: as the grinding depth increases, so does the resistance encountered by the abrasive wheel, which in turn causes the motor to draw more current and produce higher torque. By establishing a strong positive correlation between material removal rate and motor torque, the team effectively turns the motor itself into a durable, high-bandwidth sensor—capable of operating reliably in environments where delicate force transducers would quickly degrade due to vibration and shock.
The innovation lies not just in the sensing mechanism but in how the feedback is used. Instead of modulating motor speed or applying external force compensation, the control system uses torque deviation—the difference between a predefined target torque and the actual measured value—to generate corrective commands for the robot’s end-effector position. This closed-loop architecture enables continuous, millisecond-level adjustments to the TCP along the Z-axis, ensuring the grinding tool maintains optimal contact pressure regardless of surface irregularities.
At the heart of the system is a PI controller implemented on an S7-200SMART PLC, chosen for its industrial ruggedness and deterministic processing capabilities. The selection of a programmable logic controller over a general-purpose PC reflects a deliberate design choice: reliability in electromagnetically noisy factory floors. PLCs are known for their immunity to interference, predictable scan cycles, and proven track record in mission-critical applications—qualities essential for maintaining control stability during prolonged grinding operations.
The control logic operates in discrete time steps. At each sampling interval, the actual motor torque—measured via analog voltage output from the servo drive—is compared against a user-defined setpoint. This error signal feeds into the PI algorithm, which computes a correction factor. Rather than directly commanding a motor response, this output is mapped to a positional adjustment for the robot’s TCP. The adjusted coordinate is then transmitted to the ABB IRB6700 industrial robot via a serial communication interface, where it modifies the active tool frame in real time.
This approach circumvents the need to alter the original programmed path, preserving the integrity of the master trajectory while allowing for dynamic local corrections. It also avoids the computational overhead associated with full kinematic recalculations or online path replanning. By adjusting the tool coordinate system rather than the path points themselves, the method achieves rapid response without introducing latency or risking trajectory discontinuities.
One of the most compelling aspects of the research is its empirical grounding. The team did not rely solely on theoretical models or simulations. Instead, they conducted extensive physical testing on actual cast iron workpieces with highly irregular surfaces, including regions with pronounced convex features that would typically cause either excessive material removal or tool lift-off in open-loop systems.
To validate their approach, the researchers designed a rigorous comparative experiment. Three scenarios were tested: a baseline condition with no real-time adjustment (Control Group 1), a version using only proportional (P) control for TCP correction (Control Group 2), and the full PI-controlled system (Experimental Group). The results were striking.
In the unadjusted scenario, torque readings fluctuated wildly—ranging from near-zero (indicating no contact) to over 1.7V (signaling excessive loading). Such variability would inevitably lead to inconsistent surface finishes, premature tool wear, and potential damage to both the workpiece and the robot. In the P-control case, while some stabilization was achieved, the system exhibited persistent oscillations around the target value, never settling into a steady state. This residual instability stemmed from the inability of a pure P controller to eliminate steady-state error—a well-known limitation in control theory.
Only the PI-controlled system demonstrated the desired behavior: rapid convergence to the target torque, minimal overshoot, and sustained stability throughout the grinding cycle. Statistical analysis revealed that the experimental group maintained torque within ±0.05V of the mean 69.9% of the time—more than double the performance of the P-only system and vastly superior to the uncontrolled case. The average torque deviation was reduced to just 0.103V, translating into a material removal accuracy of ±0.02mm against a target depth of 0.50mm—an error margin under 10%, well within acceptable industrial tolerances.
Post-process inspection of the finished surfaces provided visual confirmation of the algorithm’s effectiveness. The uncontrolled and P-controlled samples showed clear signs of uneven grinding: patches of untouched material adjacent to over-polished zones, resulting in a visibly wavy topography. In contrast, the PI-controlled sample exhibited a uniform, lustrous finish across the entire surface, with no discernible variation even over the previously problematic convex regions.
The decision to omit the derivative (D) term from the controller was both pragmatic and insightful. While PID controllers are often considered the gold standard in feedback systems, the derivative component amplifies high-frequency noise—a major concern in grinding applications where mechanical vibrations can reach several hundred hertz. Field tests confirmed this suspicion: introducing even a small derivative gain caused severe oscillations, destabilizing the entire system. By opting for a PI configuration, the team prioritized robustness over theoretical optimality, a hallmark of good engineering judgment.
Tuning the controller parameters was accomplished through an iterative experimental process, as precise modeling of the grinding dynamics proved impractical given the stochastic nature of cast surface profiles and the nonlinear friction characteristics of abrasive contact. Starting with conservative values, the researchers systematically varied the proportional gain (Kc) and integral time (Ki), observing system response under real operating conditions. They found that a proportional gain of 1.0 and an integral time of 50 provided the best balance between responsiveness and stability. Lower integral times led to oscillation; higher values caused sluggish correction and drift.
An important contribution of the study is the establishment of a quantitative relationship between grinding depth and motor torque. Through calibration tests, the team generated a lookup table mapping specific voltage outputs (representing torque levels) to expected material removal rates. For instance, a torque signal of 1.00V corresponded to a grinding depth of 0.75mm, while 0.65V aligned with 0.50mm. This empirical model allows operators to set precise processing parameters based on desired finish quality, enabling predictable and repeatable outcomes across batches.
The implications of this work extend beyond the immediate application of cast part finishing. The core idea—using motor torque as a proxy for process force and feeding it back into geometric control—could be adapted to other robotic machining tasks such as deburring, polishing, or milling, especially in environments where traditional force sensing is impractical. The use of a PLC as the control brain also opens doors for integration into existing factory automation ecosystems, where Siemens, Allen-Bradley, and similar platforms already manage production lines.
Moreover, the approach enhances the autonomy of robotic cells. By enabling real-time adaptation to unknown surface geometries, it reduces the need for precise pre-scanning or offline programming, lowering setup times and increasing flexibility. This is particularly valuable in job shops or low-volume/high-mix manufacturing settings, where rapid changeovers are essential.
From a safety and maintenance perspective, the system offers additional benefits. Preventing motor overloads reduces thermal stress on the drive system, extending component life. Avoiding dry runs (where the wheel spins without contact) minimizes unnecessary wear on expensive abrasive tools. And by maintaining consistent contact force, the risk of sudden tool breakage or workpiece ejection is significantly reduced.
The research also highlights a growing trend in industrial robotics: the shift from purely kinematic programming to sensor-informed, closed-loop control. As manufacturers demand higher precision and tighter tolerances, open-loop systems are becoming obsolete. The future belongs to robots that can perceive, reason, and adapt—what some call “cognitive automation.” This study represents a practical step in that direction, demonstrating that even relatively simple control architectures, when thoughtfully designed, can yield transformative results.
Another noteworthy aspect is the choice of communication protocol. Rather than relying on high-speed Ethernet or proprietary fieldbuses, the team implemented a serial RS232 link between the PLC and the robot controller. This decision underscores the importance of reliability over bandwidth in industrial settings. Serial communication, though slower, is less susceptible to packet loss and timing jitter, ensuring that critical TCP adjustment commands are delivered without corruption. The handshake mechanism—where the robot requests data and the PLC responds—further enhances determinism, preventing data collisions and ensuring synchronization.
The success of this project also speaks to the value of interdisciplinary collaboration. It combines principles from control theory, mechanical design, electrical engineering, and industrial robotics. The researchers had to understand not only how PI controllers behave but also the mechanics of abrasive contact, the characteristics of servo motors, the programming environment of ABB robots, and the real-time constraints of PLC execution. Such holistic expertise is increasingly necessary in modern engineering, where systems are rarely siloed.
Looking ahead, several avenues for improvement and extension suggest themselves. One is the incorporation of multi-axis compensation. While the current system adjusts only the Z-axis (normal to the surface), future versions could apply corrections in X and Y directions as well, compensating for lateral deviations or angular misalignment. Another possibility is adaptive parameter tuning, where the controller gains are adjusted on the fly based on material type or wheel wear. Machine learning techniques could also be employed to predict surface topography and preemptively adjust the path.
Additionally, the method could be combined with vision-based inspection systems to create a fully autonomous grinding cell. After processing, a camera could scan the surface, compare it to a CAD model, and feed the error back into the next iteration—closing the loop not just during but also between operations.
Despite its many strengths, the system is not without limitations. It assumes a monotonic relationship between torque and depth, which may not hold under all conditions—such as when grinding heterogeneous materials or when the abrasive wheel becomes clogged. It also requires careful calibration for each tool-workpiece combination, which could be time-consuming in highly variable production environments. Furthermore, the reliance on a single-axis correction limits its effectiveness on complex 3D surfaces with rapidly changing normals.
Nevertheless, the overall contribution is significant. The paper demonstrates that sophisticated process control does not necessarily require expensive sensors or complex algorithms. Sometimes, the most elegant solutions are those that work with the constraints of the environment, using available signals in clever ways. By reimagining the servo motor not just as an actuator but as a sensor, Du Longfei and Wang Yu have opened a new pathway for intelligent robotic machining.
As global manufacturing continues its push toward Industry 4.0, solutions like this will become increasingly vital. The ability to maintain consistent quality in the face of variability—whether in raw materials, environmental conditions, or equipment wear—is what separates automated systems from truly smart ones. This research, grounded in practical application and validated through real-world testing, exemplifies the kind of innovation that drives progress in industrial robotics.
The methodology’s compatibility with widely available hardware—ABB robots, Siemens PLCs, Delta servos—ensures that it can be readily adopted by manufacturers without requiring massive capital investment. Its simplicity, robustness, and effectiveness make it a compelling option for anyone seeking to improve the consistency and reliability of robotic surface finishing operations.
In an era where artificial intelligence and digital twins dominate the discourse, this work serves as a reminder that fundamental engineering principles—feedback, control, and adaptation—remain at the core of technological advancement. It is a testament to the enduring power of classical control theory, applied with ingenuity to a modern industrial challenge.
The implications for productivity, quality, and worker safety are clear. By automating what was once a highly skilled manual task—adjusting pressure by hand based on sound and feel—the system reduces human error, frees operators for higher-value activities, and creates a safer workplace by minimizing exposure to dust, noise, and moving machinery.
As robotics continues to permeate heavy industry, studies like this one will shape how machines interact with the physical world—not as rigid executors of pre-defined commands, but as responsive, adaptive agents capable of intelligent behavior. The future of manufacturing is not just automated; it is perceptive, resilient, and self-correcting. And this research brings us one step closer to that reality.
Research on iso-thickness grinding by robot based on PI control by Du Longfei and Wang Yu, School of Mechanical and Electrical Engineering, Qingdao University, published in China Mechanical Engineering, DOI: 10.16731/j.cnki.1671-3133.2021.01.005