Diagnostics of Mobile Work Machines Using Dynamic Mathematical

Computational methods
in mechanical
engineering product
development
Diagnostics of Mobile Work Machines Using Dynamic
Mathematical Models and Joint Probability Distributions
Petteri Multanen, Tomi Krogerus, Mika Hyvönen and Jukka-Pekka Hietala
Tampere University of Technology / Dept. of Intelligent Hydraulics and Automation (IHA)
INTRODUCTION AND GOALS
 Diagnostics of mobile work machines is challenged by:
• Limited amount of sensors due to relatively low cost of machines.
• Harsh and highly varying operating conditions and operators.
• General problematic related to data analysis and reasoning.
 Simulation models and simulators are already necessity
in the development of highly automated machines.
 Project goal was to develop method for using
HIL simulators to support the diagnostics of machines.
 Tools and algorithms were developed for feature recognition
and machine state analysis, and were tested with real machines.
DYNAMIC MATHEMATICAL MODELS AND
HIL SIMULATOR
B
• Hydrostatic drive
Ip
• Work hydraulics and
In
TI
fluid characteristics
M
n
I
• Diesel engine
n
I
ε
• Mechanics, machine body
I
Ip
and tyre-road interaction
A
• Models were verified with
several lab and
Hydrostatic drive of wheel loader
and jammed flush valve:
field measurements
n
n
I
I
Examples of autonomous mobile work machines used for
the testing of machine diagnostics.
MAINTENANCE PROCEDURE UTILIZING HIL SIMULATORS
HIL simulator platform at IHA.
Analysis of time series data using joint probability method.
Pearson´s correlation coefficients for data sets xi and xk
EXPERIMENTS
 Wheel loader was used for the testing of data analysis.
 Recorded control references of real machine were inputs
for the mathematical models in HIL simulator.
 Only 4 variables – Diesel engine rotational speed, Pump
displacement, Pressure A and Pressure B.
 41 test drives; 20 drives were used in the training phase,
i.e. for statistical model generation, and
21 drives were used in
the actual testing phase.
 Machine fault was
a jammed flushing valve
in hydrostatic transmission.
Joint probability of
coefficients in
one data segment.
Example of drive path in simulator.
Logarithm of joint probability
EXPERIMENT RESULTS
-40
Simulated vs Real undamaged (Train)
Simulated vs Real undamaged (Test)
Simulated vs Real damaged (Test)
-50
Mean(undamaged train) = -55.00
Mean(undamaged test) = -55.97
-60
Mean(damaged test) = -60.14
Threshold = -70
-70
-80
0
50
100
150
Segments
Joint probability distributions of training and testing data.
CONCLUSIONS
• This poster presents method for using simulators for the diagnostics of machines
and analysis tools for the recognition of machine condition.
• In joint probability distribution method the probabilities of multiple correlation
coefficients are compared instead of comparing correlations directly.
This enables the detection of anomalies, rare situations with low probabilities,
from which one can conclude if there is something wrong in the system.
• Analysis method was applied to the diagnostic of mobile work machines and was
tested by producing a jammed flush valve to the hydrostatic drive of wheel loader.
• Test results show clearly lower probabilities for test drives where fault is present.
• Analysing methodology enables the detection of sudden critical faults as well as
slowly evolving faults.
• Simultaneous examination of several variables enables also a more generic
approach of detecting several different anomalies and applying it to
different machine types.
This poster is based on Tampere Univ. of Technology´s Sub Project 2 in SIMPRO:
Utilization of simulation data to support the maintenance of mobile work machines
CONTACT
Petteri Multanen
Tampere University of Technology
Dept. of Intelligent Hydraulics and Automation
Tel: +358 50 599 4329
Email: [email protected]
Publications:
• Multanen, P. and Hyvönen, M. Utilization of R&D Simulators to Support the Maintenance of Mobile Work
Machines. Proc. of 8th international conference on condition monitoring and machinery failure
prevention technologies, Cardiff, UK, 20-22 June, 2011, pp. 1-12.
• Hietala, J-P., Krogerus, T., Multanen, P., Hyvönen, M. and Huhtala, K. Novel Procedure for Supporting
Maintenance of Mobile Work Machines Using R&D Simulators. Proc. of CM 2014 and MFPT 2014
conference. The 11th International Conference on Condition Monitoring and Machinery Failure
Prevention Technologies, 10-12 June 2014, Manchester, UK. 9 p.
• Krogerus, T., Hyvönen, M., Backas, J. and Huhtala, K. Anomaly Detection and Diagnostics of a Wheel
Loader Using Dynamic Mathematical Model and Joint Probability Distributions. The 14th Scandinavian
International Conference on Fluid Power. May 20-22, 2015, Tampere, Finland. 14 p.
• Krogerus, T. et al. Joint probability distributions of correlation coefficients in the diagnostics of
mobile work machines. IMechE Journal part C: Journal of Mechanical Engineering Science. In review.