NUST Institutions Library Catalogue

Reliability of high precision linear motion systems is one of the main concerns in
industrial and military systems. The performance and repeatability of these systems are
influenced by their respective Ball Screw (BS) linear drives. A fault in these members severely
affects positioning accuracy and safe working of overall system. BS linear drives perform flight /
application critical job and are responsible to provide precise linear motion while carrying thrust
loading. BS drives are specifically designed on the basis of desired operational parameters like
power rating, drive torque, slew rate, efficiency, friction, and mechanical backlash etc. These
operating parameters significantly affect the functional performance of linear electro-mechanical
systems. At present, few techniques are available to monitor BS drives for aerospace and
industrial systems.
This research works to improve reliability of ball screw drive linear systems by modeling
and monitoring the performance factors through analytical redundancy and intelligent deep
learning. In the past, some traditional techniques have been employed to address these problems;
however these techniques show limitations like insufficient data acquisition, requirement of
dedicated model developer and poor domain adaptation. Recently, deep learning techniques have
been introduced and are becoming more popular to detect and characterize various fault signal
analysis problems due to their robustness and accuracy.
The aim of this research is to provide solution of mechanical faults identification and
classification problems for BS linear drives. A fault diagnostic algorithm is designed based on
dynamic mathematical model and a remnant filter is implemented to detect signal errors. The
remnant filter generates residual signal proportional to the error induced. Fault detection
thresholds are set and decision logic is established based on position measurement corrections to
compare residual signal with the lower and upper pre-defined threshold constants.
Accuracy in faults identification is highly dependent on improved features extraction. For
this purpose, a novel Residual Twin CNN (ResT-CNN) is proposed that uses combination of 1-D
and 2-D CNN in parallel learning which improves features extraction performance; followed by
knowledge base-Remnant-PCA (Kb-Rem-PCA) architecture in combination with multi-class
support vector machine (Mc-SVM). Current and Position signal data was collected under
different load domains. This novel hybrid combination proved very effective in accurate faults
identification and classification.
The performance of proposed intelligent technique was successfully tested and validated
on different datasets including IMS-UC (Intelligent Maintenance Systems – University of
Cincinnati) publically published bearing dataset, Paderborn published multi-stage bearing
dataset, Current signal dataset for multiple fault modes of BS drive and Position measurement
data for multi faults cases for BS linear drive.
The actual Signal and Model Fit Simulated Data for BSD system was compared. The
testing results proved the effectiveness and superiority of proposed model against different state
of the art techniques. The proposed novel framework was also tested for system's stability under
different load domains. The results reveal highly competitive values greater than 95% in terms of
accuracy and precision for different faults cases