Catching Gripper Wear Earlier: An Information Monitor Compared with Control Charts in a Simulated Pick-and-Place Cell

In the quiet hum of a modern factory, industrial robots move with a precision that humans cannot match, lifting and placing thousands of parts every hour. These machines rely on grippers, mechanical hands that close to grab an object and release it to move it along the line. For years, engineers have known that these grippers do not fail suddenly; they wear down slowly, becoming slightly less steady over time. The challenge has always been knowing when to stop the machine for maintenance. If a team waits too long, the gripper becomes so loose that the parts it holds fall out of alignment, creating a pile of scrap. If they stop too early, they waste time and money on unnecessary repairs. The goal is to catch the wear at the very first sign, long before the parts themselves are ruined, giving the maintenance crew a window of time to act while the machine is still producing good work.

To find this early warning, a team of researchers led by Wael Hafez, Amirhossein Nazeri, and Chenan Wei set up a virtual test environment. They did not use a physical robot in a real factory, but instead created a detailed computer simulation of a pick-and-place cell. In this digital world, they could control every variable perfectly. They programmed a robot to run a single task repeatedly: closing its gripper on a thirty-millimeter part, lifting it, and placing it down. They then introduced a specific type of wear into the simulation, not by breaking a part, but by adding a small amount of random jitter to the gripper's movement. This jitter mimics the way a real mechanical joint might start to shake or wobble as its bearings degrade. The researchers wanted to see if a new type of monitoring system could spot this tiny wobble before the standard tools used in factories could.

The standard tools are known as control charts. These are graphs that factory engineers have used for decades to watch a single number, such as the average position of the gripper or how much that position varies from one cycle to the next. When the number drifts too far from the normal range, the chart raises an alarm. The researchers tested two common types of these charts, which look for slow shifts in the average or the spread of the data. They compared these against a new method called an information monitor. Instead of just watching one number, this new system looks at the relationship between three things: the command the robot was told to execute, the actual movement it made, and the movement it made a moment later. It asks a simple question: does the same command still produce the same response? If the relationship between the command and the response changes, it suggests the machine is wearing out, even if the average movement looks normal.

The results of the simulation were striking. The researchers ran the robot through nearly 288,000 production cycles, which equates to about 160 hours of continuous operation. They introduced wear gradually, starting at a level just slightly above normal and increasing it until the gripper was moving with four times its usual amount of jitter. The new information monitor sounded the alarm on every single machine as soon as the jitter reached between 1.2 and 1.45 times the normal level. At this point, the gripper was still holding parts perfectly within the required limits. The machine was still making good products. In contrast, the standard control charts remained silent. They did not raise an alarm until the jitter had grown to about three times the normal level. By the time the charts finally screamed for help, the gripper was so loose that nearly every part being produced was already defective.

This difference in timing is crucial for factory operations. The researchers found that the first bad parts appeared when the jitter reached 1.5 times the normal level. The information monitor caught the problem before that threshold was ever crossed, giving the maintenance team a clear warning while the parts were still good. The standard charts, however, only detected the problem after the damage was already done. In the simulation, the new monitor caught the wear on every machine at the lowest levels tested, while the traditional charts missed it entirely until the machine was already failing. Furthermore, when the researchers tested the system on eighty machines that were running perfectly healthy, neither the new monitor nor the old charts raised a single false alarm. This suggests that the new method is sensitive enough to find the earliest signs of trouble without causing unnecessary panic.

The study also revealed that the new monitor is very sensitive to changes in the job itself. If the researchers changed the size of the part, sped up the cycle, or shifted the holding position, the monitor sounded an alarm just as loudly as it did for wear. This means the system cannot simply be left running on any machine forever. It requires a fresh "healthy baseline" to be recorded every time the job changes or after maintenance is performed. If the part size changes, the monitor needs to learn what the new normal looks like. This is not a flaw, but a necessary feature; the system is so precise that it notices any deviation from the pattern it was taught. The researchers noted that this sensitivity means the monitor is best used as a trigger for an inspection, telling a human to check the machine, rather than as a final diagnosis of what is wrong.

While the results are compelling, the researchers are careful to note that this work took place entirely in a computer simulation. The wear they introduced was a steady, predictable increase in random movement, which is a specific type of failure. Real-world robots face sudden shocks, intermittent electrical issues, and complex physical interactions that were not part of this test. The simulation also used a simplified model that did not include gravity or the weight of the parts, focusing only on the ability to compare different monitoring methods. Despite these limitations, the findings offer a clear path forward. The study suggests that for factories running a fixed job, relying only on the average movement of a robot is not enough. Engineers should also track how much the movement varies, and they should consider using methods that look at the relationship between commands and responses. By doing so, they might be able to catch the first whisper of wear, schedule a repair during a planned break, and keep the production line running without ever producing a single bad part.