Ball Screw Test Bench Dataset
In order to develop predictive maintenance algorithms to determine the RUL (Remaining Useful Lifetime) of ball screws a solid, high quality data base is crucial. To build this data base a test bench has been built to collect data under laboratory conditions. This dataset contains the data from 9 full test cycles (new ball screw until breakdown).
Each dataset contains 3 different types of data
1. Calculated mean values (Mean_XXXX_XX_XX_XX_XX_XX)
this values are calculated as mean from 1,5s time interval when the speed is constant – there are always 2 files every 200 cycles (one for each direction of movement)
- Number of executed cycles
- Stiffness
- 3x temperature (3 ball screw nuts)
- Torque both motors
- For each of 5 accelerometers (2 on ball screw, 3 on bearing housings):
- RMS value of the velocity spectrum from 10 Hz to 1kHz as defined in the ISO 10816 [mm/s]
- RMS Value of the velocity raw signal from 1 Hz to 10kHz [mm/s]
- Absolute maximum value of the raw signal from 0 Hz to 10kHz [mg]
- Crest-Factor (Ratio of the absolute maximum value to the RMS value) of the raw signal from 0 Hz to 10kHz
- Kurtosis (4th statistical moment) of the raw signal from 0 Hz to 10kHz
- Skewnes (3rd statistical moment) of the raw signal from 0 Hz to 10kHz
- 6 frequency bands
2. Raw data from whole cycle (Raw_2022_07_29_01_31_45)
- Limitation – sampling frequency only 500Hz
- Good for watching torque
- Good for watching abnormalities at the end of the trajectory
3. Raw data from whole cycle (XXXX_yDataXXXXX, XXXX_xDataXXXXX, XXXX_yFFTXXXXX, XXXX_xFFTXXXXX
- Sampling frequency 25,78kHz
- Shor time period during constant speed running
- Raw data (x and y values) and FFT spectrum calculated from this data
Due to time constraints (project duration only 15 months), it was decided to test smaller ball screws because smaller dimensions mean a lower maximum load and also a shorter lifetime. This also ensured that the demonstration was financially economical, as compared to larger ball screws, there are lower costs both for the purchase of the ball screws themselves and for the design and construction of the demonstrator, which can be sized for the lower load. After consultation with the manufacturer, KSK Precise Motion, Inc., a 20 mm diameter ball screw was selected. At the same time, a load cycle was designed by the manufacturer. For an axial load force of 4000 N, 1000 rpm, and a symmetrical test cycle, a lifetime of approximately 720 h was calculated. The test cycle was programmed according to the manufacturer's design. First, the ball screw under test is loaded with torque from the servomotor, with the braking servomotor holding the ball screw nuts in the same position. Once the desired load is achieved, the braking servomotor adjusts the braking torque so that the connecting plate will move to the other side of the demonstrator at the given acceleration and speed. Here, the orientation of the load torque is changed on the drive motor (e.g. from 3.185 Nm to - 3.185 Nm), and when this is reached, the braking motor is used to perform a backward movement.
As already mentioned, at the end of the service life, the nut loses tension and backlash occurs. The nut preload cannot be measured directly - when converting the preload force to pressure in the spacer ring, the pressure value was found to be too small and not within the measurable range of any sensors (e.g. strain gauges). However, the preload is related to the ball screw stiffness, which can be measured indirectly, as well as the clearance (or positioning accuracy). Based on previous experience, it was decided to use primarily vibration analysis, supplemented by information from temperature sensors and motors (torque). Ultrasonic sensors are currently a less explored area and were only developed during the project. Conversely, microphones were not used at all, as they work on the same principle as vibration analysis (sound is vibration in the air), but due to the high environmental disturbance, they will warn of impending end-of-life much later. Tribometers were not used because of the complexity of measurement and evaluation - the ball screws were lubricated with grease, not oil. Although the number of sensors used is high (11 sensors in total), in practice this number is not necessary. When monitoring the lifetime of a ball screw in real operation, it is sufficient to use two accelerometers in the radial direction, measuring in mutually perpendicular directions (similar to those on the demonstrator ball screw nut). The other accelerometers on the bearing houses were used as additional measurements in a case bearing degradation and failure occurred. In that case, the largest increase in vibration will be on these accelerometers. Capacitive position sensors are not needed in actual operation, because with predictive maintenance algorithms, we will be able to determine the condition of the ball screw using vibration. There is no need for temperature sensors because, as we already know, deterioration of the CS will be reflected in the vibration signal long before the temperature increase that occurs at the very end of the lifetime