How the information monitor differs from control charts: control charts watch one average or spread and alarmed at about 3x normal variation when nearly all parts were bad; the information monitor checks whether command, response and next response still fit together and alarmed at 1.2-1.45x, before the first bad parts.

Wear caught before the first bad parts

In a simulated pick-and-place cell, the information monitor alarmed on every machine at 1.2 to 1.45 times normal variation, before the first bad parts appeared. CUSUM and EWMA charts on spread alarmed only at about 3 times, when nearly all parts were already bad; on the mean they never alarmed. No method raised a false alarm on 80 healthy machines. The study covered 287,600 cycles, about 160 hours of simulated production; it is not a test on a physical robot.

Gripper condition over its working life: the information monitor alarmed at 1.2-1.45x normal variation while parts were still good; EWMA and CUSUM control charts alarmed at about 3x, when nearly all parts were bad. Simulated pick-and-place cell.

How it works

Your controller already records what it commanded and what happened next, every cycle. We read how command and response stay related. When wear, friction or play changes that relationship, the reading moves, even while averages and spreads still look normal. We don't sell a finished product: we provide the method and engineering support so your team can build information monitors into its own maintenance systems.

Catch wear before it makes bad parts

An information monitor for machines that log a command and a response. It checks whether the same command still gives the same response, using data your controller already records. No training, no machine model. Blind test: you send logs with wear or faults you already know about, unlabeled; we return what we detect and when; you score us against your own records.

Request a blind test

Test it on your own data

Send us logs. Normal cycles from a healthy machine, plus a log containing wear or faults you know about but don't label.
We analyse blind. We return the timestamps and severity of what we detect.
You score us. You hold the answers; we see them only after we submit ours.

Is this for you?

These are examples: any machine that logs a planned command and a measured response can be tested. Tested so far: robot grippers (simulation) and robot arm joints (real-arm benchmark, see Robot Monitoring). Servo axes, spindles, conveyors, pumps and other actuators follow the same principle but are not yet tested.

What to send: two files, a normal baseline from a machine in good condition (about 80 minutes of production is enough; fewer is fine to start) and a test log with wear or faults you know about, labels removed. Core columns: planned and measured position (or command and response), with a timestamp and cycle ID. A change of part, program, speed, payload or tool needs a new baseline.

See full data guide and request a test

How Much Model Does Condition Monitoring Need? Reading a Six-Axis Arm from Its Control Loop. (2026), Research Square, doi.org/10.21203/rs.3.rs-11101227/v1
Catching Gripper Wear Earlier: An Information Monitor Compared with Control Charts in a Simulated Pick-and-Place Cell. (2026). Preprint available on request.