Fault-Tolerant Control Strategy for Robot-Joint PMSM Drives Under Encoder Faults

Robots that move with the precision of a human hand rely on motors that are incredibly fast and responsive. Inside the joints of these machines, permanent magnet synchronous motors act as the muscles, turning electrical energy into smooth, controlled motion. To keep these motors moving exactly where they should, the system needs to know the precise position of the spinning rotor at every fraction of a second. Usually, this information comes from a small device called an encoder, which acts like a digital eye, watching the shaft and sending a constant stream of position data back to the computer brain. If this eye goes blind or starts seeing things that aren't there, the motor can lose its way, leading to jerky movements or a complete shutdown. In the delicate world of robotics, where a robot might be assembling a microchip or assisting in surgery, losing that position feedback is not just an inconvenience; it is a failure that can halt production or cause damage.

The challenge for engineers is that these motors operate across a vast range of speeds, from a complete standstill to thousands of revolutions per minute. At a standstill, the motor is quiet and produces no natural electrical signals that a computer can easily track. At high speeds, it generates strong signals, but the methods used to track it at low speeds become noisy and unreliable. Traditionally, if the encoder fails, the robot stops. However, a team of researchers at Jiangsu Normal University has developed a way to keep the motor running even when its primary "eye" fails. They created a system that uses two different mathematical techniques to guess the rotor's position, fusing them together so that if one method is weak, the other takes over. This allows the robot to continue its work seamlessly, even after the encoder stops working correctly.

The researchers focused on three specific ways an encoder can fail: a sudden jump in the reported angle, the output getting stuck on a single value, or a loss of pulses that makes the speed reading erratic. To handle these failures, they designed a control strategy that runs two parallel estimation systems in the background while the encoder is still working. One system, known as high-frequency injection, works by sending a tiny, rapid electrical signal into the motor and listening to how the motor's internal magnetic structure reacts. This method is excellent for finding the position when the motor is not moving or moving very slowly. The second system, called a sliding-mode observer, looks at the voltage and current flowing through the motor to calculate the position based on the motor's own physics. This method is highly accurate when the motor is spinning fast.

The clever part of this new strategy is how it manages these two systems. Instead of simply switching from one to the other as the speed changes, the researchers created a smooth blend. They determined exactly how much trust to place in the low-speed method versus the high-speed method at every single speed, and they adjusted the strength of the high-frequency signal independently to keep the motor efficient. While the encoder is healthy, the computer constantly compares the encoder's report with the combined guess from these two systems. If the encoder suddenly starts reporting a position that is wildly different from what the two systems agree on, the computer knows something is wrong. It does not panic immediately; it waits for a short period to confirm the fault, ensuring it does not mistake a momentary glitch for a real failure.

Once the fault is confirmed, the system performs a delicate handover. It instantly disconnects the faulty encoder from the control loop and switches to using the combined estimate from the two mathematical systems. To prevent the robot from jerking during this switch, the system applies a temporary compensation that fades out over just 50 milliseconds, smoothing the transition so the motor continues to turn without a hitch. The researchers tested this approach extensively using computer simulations and a physical test platform with a real motor and controller. In the simulations, they introduced various faults while the motor was running at different speeds, from a slow 100 revolutions per minute to a fast 1500 revolutions per minute. In every case, the system detected the fault within 16 to 31 milliseconds and successfully maintained closed-loop control. The error in the motor's angle after the takeover was remarkably small, staying within a range of less than 1.5 degrees, which is barely perceptible in a mechanical system.

The physical experiments on the test platform confirmed that this strategy works in the real world, not just in a computer model. The team ran the motor from a complete stop up to 3000 revolutions per minute, testing how well the two estimation methods worked together across the entire speed range. They simulated a stuck encoder at zero speed, a sudden angle jump at 800 revolutions per minute, and a pulse-loss fault at 3000 revolutions per minute. In each scenario, the system identified the problem, isolated the broken encoder, and kept the motor running smoothly using the estimated position. Even when the encoder was completely stuck before the motor started, the system used the high-frequency injection method to start the motor from a standstill without any external help. The results showed that the motor could recover from these faults with minimal disturbance to its speed or the electrical current flowing through it.

This work demonstrates that robot joints do not have to be fragile, dependent on a single sensor that, if broken, brings the whole machine to a halt. By fusing different ways of estimating position and creating a robust way to switch between them, the researchers have shown that motors can be made resilient to sensor failures. The approach allows for a continuous, smooth operation even when the primary feedback device fails, ensuring that the robot can finish its task without interruption. While the study focused on specific types of encoder faults and used a controlled test environment, the results provide a clear path toward more reliable and fault-tolerant robotic systems that can operate safely in complex and demanding environments.