Torque ripple suppression in switched reluctance motors using iterative learning control to optimize the torque sharing function

Electric vehicles rely on motors that are quiet, smooth, and efficient, yet one common type of motor used in these vehicles often struggles with a specific flaw: a rough, shuddering sensation at low speeds. This issue stems from the way the motor's internal parts interact. Unlike the smooth, continuous rotation of other motors, this type of machine, known as a switched reluctance motor, operates by switching electrical power on and off to different sections as a rotor turns. While these motors are prized for their rugged simplicity and lack of rare-earth materials, the act of switching power from one section to the next creates a momentary imbalance. The force that turns the motor does not transfer perfectly from one section to the next, resulting in a jerky motion known as torque ripple. For a passenger in a car, this feels like a vibration or a hesitation; for the vehicle's components, it creates unnecessary wear and noise.

To solve this, engineers have long used a method called a torque sharing function. Imagine trying to pass a heavy load from one person to another; if the first person lets go before the second person has a firm grip, the load drops. If the second person grabs too early, they might pull against the first. A torque sharing function is a set of instructions that tells the motor exactly how to split the effort between the old section and the new section during that critical handover. For years, engineers have used a standard, pre-written set of instructions based on a smooth curve to manage this transition. However, these fixed instructions assume the motor behaves perfectly, ignoring the fact that real motors are messy. Magnetic fields inside the motor do not behave in a perfectly predictable way, and the electrical currents cannot rise or fall instantly. Because of these physical realities, the standard instructions often fail to match what the motor actually needs to do, leaving the shuddering problem only partially solved.

Researchers at Henan University of Science and Technology have proposed a new approach that allows the motor to learn from its own mistakes. Instead of relying on a fixed set of instructions that never changes, they developed a system that uses a technique called iterative learning control. This method works on a simple principle: if a task is repeated over and over, the system can look at the error made in the previous attempt and adjust its behavior for the next one. In the context of the motor, which spins in a continuous loop, the system observes the difference between the smooth, ideal force it wanted to produce and the rough, actual force it generated during one full rotation. It then uses that specific difference to tweak the instructions for the very next rotation.

The researchers built a detailed computer simulation of an eight-pole motor to test this idea. They started with the standard, pre-written instructions and let the system run through hundreds of simulated rotations. With each turn, the system analyzed the torque ripple—the jerky fluctuations—and made tiny adjustments to the sharing function. It did not change the entire rulebook at once; instead, it refined the specific moments where the power switched from one section to another. Over time, the instructions evolved from a generic curve into a custom profile that perfectly matched the unique, non-ideal behavior of that specific motor. The system learned that to get a smooth ride, it needed to ask for slightly more current in one spot and slightly less in another, compensating for the magnetic delays that the standard rules ignored.

The results of these simulations were striking. When the researchers tested the motor at low speeds of 100, 200, and 400 revolutions per minute, the new learning system significantly smoothed out the motion. Compared to the traditional method, the new strategy reduced the roughness of the torque by nearly 40 percent across all tested speeds. At 100 revolutions per minute, the reduction was 39.69 percent; at 200 revolutions per minute, it was 38.86 percent; and at 400 revolutions per minute, it was 38.29 percent. The motor did not just perform better on average; the peaks and valleys of the force became much more uniform, eliminating the sharp spikes that cause vibration. The system reached this improved state after about nine simulated rotations, at which point the adjustments became so small that the performance had effectively stabilized.

While the findings are promising, the researchers are careful to note the boundaries of their work. These results come entirely from computer simulations, not from a physical motor on a test bench. The study confirms that the method works in a digital environment where variables can be controlled perfectly, but the real world introduces new challenges like friction, heat, and manufacturing imperfections that were not present in the simulation. The authors acknowledge that the system was tuned for low-speed operation, and its effectiveness at higher speeds remains to be seen. Furthermore, the study did not measure how this extra processing might affect the overall energy efficiency of the motor or the lifespan of its electronic components.

Despite these limitations, the study offers a clear path forward for improving electric vehicle comfort. By moving away from rigid, pre-programmed rules and toward a system that adapts based on real-time performance, engineers can potentially eliminate the roughness that has long plagued switched reluctance motors. The next step, as the researchers outline, is to build a physical prototype to see if the lessons learned in the computer hold up when the motor is actually spinning. If the physical tests mirror the simulation, this learning approach could become a standard tool for making electric vehicles smoother, quieter, and more reliable.