See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228943830

Rapid control prototyping in test sequence design for condition monitoring of

a hydraulic valve

Article · January 2004

CITATIONS READS

2 1,196

2 authors, including:

Vesa Hölttä
Kindhelm
24 PUBLICATIONS   87 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Vesa Hölttä on 23 April 2015.

The user has requested enhancement of the downloaded file.

 Rapid control prototyping in test sequence design
for condition monitoring of a hydraulic valve
Lauri Palmroth Vesa Hölttä
Institute of Automation and Control Control Engineering Laboratory
Tampere University of Technology Helsinki University of Technology
P.O. Box 692, FI-33101 Tampere P.O. Box 5500, FI-02015 HUT
FINLAND FINLAND
[email protected] [email protected]

Abstract – Mobile working machines require condition II. MECHANISED TIMBER HARVESTING
monitoring in order to minimise downtime and maintenance
costs. Rapid control prototyping (RCP) is a control design There are different methods for mechanised timber
method where testing of new functions is done first in harvesting differing by their philosophy and the machines
simulation environment. The simulation models are converted
using automatic code generation to a prototype controller that
needed. In North America the full-tree and tree-length
can be used in field testing. The time needed to develop new methods are common whereas in Scandinavia the cut-to-
functions is reduced significantly and consequently it is possible length method is dominant. In Finland 96 % of the volume of
to test different solutions without additional delay. This paper the timber that was harvested in 2002 was felled with a forest
presents a test sequence that can be used to monitor the harvester [1].
condition of hydraulic valves of a forest machine. The purpose Despite of the high degree of automation in a harvester,
of the test sequence is to detect the instability of proportional the human operator has the main responsibility of the
pressure reducing valves. The detection of impending and efficiency and environmental impact of timber harvesting.
multiple simultaneous faults is also possible. Discrete wavelet The control unit of the harvester has software that calculates
transform is used in the analysis.
the optimal lengths of the harvested logs in order to
maximise the timber value. However, it is the responsibility
I. INTRODUCTION
of the operator to decide the final log lengths depending on
the quality the harvested timber. Usually the operator plans
A forest harvester is used for felling, delimbing and cutting
the route for the logs to be carried to the roadside. Especially
trees. It has been designed to be operable in difficult terrain
at thinnings, the operator may also have to choose which
and in operating temperatures and other conditions that vary
trees are cut down [2]. In order to make the operator more
substantially during the year and even during one day.
able to focus on decisions only a human can make, other
Modern forest harvesters are becoming more and more
functions of the harvester should be made as automatic as
computerised and a number of sophisticated control methods
possible.
such as fuzzy control and learning systems are included in
the distributed control system of the machines. The
A. Cut-to-length harvester
development of new methods requires extensive testing,
preferably in the actual working environment, and several In the cut-to-length method the trees are felled, delimbed
approaches may need to be tested before a working one is and bucked (i.e. cut to logs) with a forest harvester (Fig. 1).
found. Therefore flexible tools that enable fast design and The harvester is equipped with measuring devices which
testing are needed. measure the length, diameter and volume of the stem.
Traditionally one of the main problems in the product Optimization algorithms choose the bucking so that the value
development of mobile working machines is the difficulty of
collecting additional data and testing new control prototypes
in normal working conditions. This is due to the extremely
harsh working environment that all additional sensors and 3
2
signal processing units placed in the machines must tolerate. 1
Implementing new control algorithms often requires revising
some of the software components of the machine’s existing
control units. As a result of this, usually only small
modifications are made at a time to a product, which has
proven its functionality in practice.
In this paper an automatic test sequence for monitoring the
condition of hydraulic valves is developed. The paper begins
with an introduction to the application area, i.e. mechanised
timber harvesting. Section III presents rapid control 4
prototyping. The background of the test case and the methods
are dealt with in section IV. Simulation and field test results
are in section V, and conclusions complete the paper. Fig. 1. Cut-to-length forest harvester.
of the stem is maximized. Once the stems are processed, a III. RAPID CONTROL PROTOTYPING
forwarder carries the logs to the roadside for further
transportation. A. Overview
A cut-to-length forest harvester can be divided into four
main parts: engine and power transmission (number 1 in Fig. Rapid control prototyping (RCP) provides a quick solution
1), cabin and controls (2), crane (3) and harvester head (4). for prototyping new functions in a mobile working machine
All the power of the machine is delivered hydraulically. The that is controlled with a complex distributed control system
diesel engine is only used to rotate the supply pumps of work [3]. Rapid control prototyping process consists of the
hydraulics and hydrostatic transmission. The supply pump of following phases:
work hydraulics delivers hydraulic power to the crane, the 1. Modelling of the plant in simulation environment
harvester head and all the other auxiliary functions of the 2. Model validation
machine. The hydrostatic transmission consists of a variable 3. Designing a prototype version of the control system
displacement pump, a variable displacement hydraulic motor 4. Testing the prototype version of the control system
and mechanical transmission to wheels. against the simulation model
The cabin is equipped with the controls that are needed for 5. Testing the prototype version of the control system in
operating the functions of the harvester and with the the real plant
computer display module, which gives the operator
information on the harvesting process and on the state of the The basic idea is to develop and test the new control
harvester. strategies and algorithms in a simulation environment. As
The different parts of the harvester are interconnected via a soon as the controller design is complete and the simulation
Controller Area Network (CAN) bus. The bus enables two- results are acceptable, the controller can be translated to a
way digital communication between the harvester head, real-time prototype controller, which is used for testing the
crane and other modules. algorithm in real working conditions. The translation from
the off-line simulation environment to a real-time controller
B. Harvester head is done automatically without any handwritten code.
Two basic RCP approaches can be distinguished: fullpass
The most complex part of the harvester is the harvester and bypass [3]. With fullpass, the function prototyping
head. Its main functions are sawing, feeding, delimbing of hardware completely replaces the production controller. All
branches and measuring log length and diameter profile. sensors and actuators are connected to this electronic
Trees are felled and the stems are cut to logs with a hardware, and it has full authority to control the plant. In this
hydraulically actuated chain saw. Once sawing is complete application the bypass-method is used. It means that the
the stem needs to be fed to a new cutting point. This is done prototype controller of the new function is implemented by
with feeding rollers, which are rotated with hydraulic motors. partially revising the existing control units of the machine.
To prevent the feeding rollers from sliding, the rollers are The original control unit executes all the functions that will
pressed hard against the stem with a hydraulic cylinder. In remain unchanged, while the new algorithms are calculated
front and behind of the feeding rollers there are delimbing in a specific control unit, which also provides the additional
knives, which are pressed around the stem with hydraulic inputs and outputs. If needed, a PC provides the user
cylinders. As the stem is fed to the next cutting point, interface for the new control system. (Fig. 3)
branches are cut when they meet the delimbing knives.
Delimbing knives also prevent the stem from falling out of B. Benefits and drawbacks
harvester head grasp during the feeding operation. The main
parts of harvester head are shown in Fig. 2. The performance The RCP approach has many advantages. The use of
and functionality of harvester head have a large-scale effect simulation models helps the control designer to achieve a
on the overall timber harvesting efficiency of the machine deeper understanding of the process. Simulation experiments
and the quality of the harvested timber. may also reveal important state variables that are impossible
CONTROL UNIT AND VALVE BLOCK
to measure from the real process.

SAW UNIT

FEEDING ROLLERS

DELIMBING KNIVES

Fig. 2. Main parts of the harvester head. Fig. 3. Bypass-based rapid control prototyping. [4]
 In simulation environment it is also possible to experiment application without the user interface. ControlDesk has
with features that are not available in the hardware currently several instruments for real-time adjustment of the model and
used in the product. These features include e.g. higher controller parameters and for monitoring of the
sampling rates and larger computational power. Thus it is measurements. It is possible to record and store all the
possible to evaluate the benefits that would be obtained if the measurements and state variables to MATLAB data files
hardware in the product was upgraded to a better and from the ControlDesk. In addition to Rapid Control
presumably more expensive version. Prototyping, Simulink and dSPACE hardware provide also a
The software and hardware used in the RCP process serve Hardware-In-the-Loop (HIL) environment for testing of the
also as high performance general purpose data acquisition control modules of the production machine.
tools in addition to providing a real-time control system for
the new algorithms. Time and money are saved when the IV. TEST CASE
costly and time-consuming I/O-wiring of the machine is done
only once. The same equipment is used for testing of the new A. Background
control prototype in the real working environment as well as
for data acquisition in modelling and model validation The traditional approach in maintenance of machinery is to
process. Control prototyping does not require any perform service operations at regular time or machine hour
modifications of the machine’s existing software intervals, or when a component breaks down preventing
components. normal operation. In contrast to this, in condition based
Automatic code generation eases significantly the maintenance the decisions regarding service actions are
iterations in the control system design, because the time- based on the actual condition of the machine or a component
consuming manual coding phase of the algorithms is left out. [5].
It also removes the programming errors generated during the The primary goal of condition based maintenance is to
manual coding phase. As a whole, the time needed for decrease unexpected downtime and maintenance costs of the
designing and implementing a control system is shortened. machines. If machines are run until they break, the resulting
Thus several different solution candidates can be prototyped unplanned service breaks are longer than planned ones, due
and tested in the real process. to the fact that spare parts and skilled personnel may not be
The use of RCP may require purchasing commercial available immediately. Especially, in the case of a forest
proprietary software and hardware to be used in the design harvester, service may often be very far away from the
and testing process. However, the cost of such products may machine. On the other hand, unnecessary, i.e. too frequent,
be small when comparing with the amount of work that maintenance increases costs and downtime. Consequently,
would be required to realize a similar test setup from scratch. knowledge of the exact condition of the equipment is needed
The software and hardware can be also reused in future for scheduling and choosing maintenance optimally [5].
projects. The aim in this test case is to design and to test a test
To be able to design a good control algorithm for a plant sequence that can be used to evaluate the condition of a
using the RCP method, the plant model must be accurate hydraulic valve. The sequence should be such that it can be
enough. Unmodelled plant dynamics, nonlinearities as well performed periodically without disturbing the normal
as unexpected measurement noise and disturbances during operation of the forest harvester. If possible, the test
testing with the real plant may result in the failure of a sequence should also be performed without any additional
control algorithm that has performed well in simulations. equipment that is not included in the normal operation of the
Should this be the case, the data obtained from the plant can machine. The results of the test sequence should be
be used to refine the model of the plant before retuning or interpreted automatically so that the evaluation of the health
redesigning the control algorithm. of the valves is obtained without expertise in hydraulics.
Simulation is used for avoiding tasks that would be time-
C. Tools consuming, expensive and impractical to do in practice. In
this test case the goals of simulation are:
In this application MATLAB with Simulink and Real- 1. Finding the optimal pressure levels and positions of
Time Workshop is used as the simulation tool. MATLAB
hydraulic components and actuators
and its toolboxes provide a high level development 2. Finding the optimal inputs of the test sequence
environment for implementing different control strategies. 3. Finding the optimal sensor locations in the system.
The hardware of the specific control unit is dSPACE
MicroAutoBox. The Real-Time Interface of dSPACE If these goals were to be achieved without simulation,
provides Simulink blocks for the MicroAutoBox I/O signals, extensive work changing the hardware to test e.g. different
which replace the plant model in the simulation model. The sensor positions would be required. In simulation testing the
Simulink model is translated through Real-Time Workshop parameters of the process can be kept constant and
and Real-Time Interface of dSPACE to a program that is disturbances are replicable, whereas in real-world testing the
loaded on the specific control unit. The user interface is fault in the process could become more severe during the
created with dSPACE ControlDesk application. The specific testing, making the results to be noncomparable.
control unit may also control the plant as a standalone
 pOUT and that it is much higher than p. Turbulent flow through an
orifice is given by
2∆p
Q = CA , (2)
ρ
where C and A are the flow coefficient and the cross-
sectional area of the orifice, ∆p the pressure drop across the
orifice and ρ the density of the fluid. The conventional
turbulent orifice flow formula (2) has an infinite derivative
pS pT when the pressure drop is zero. This can cause ODE solvers
to crash during simulation. Applying a two-regime orifice
Fig. 4. Hydraulic scheme of proportional pressure reducing valve. flow formula proposed by Ellman and Piché [7], the flow
formula is better-suited for simulation. The laminar and
turbulent flow regimes are determined by the transition
B. Condition monitoring of a hydraulic valve pressure ∆ptr. The turbulent flow (∆p > ∆ptr) is given by

In this test case condition monitoring is applied to a ∆p

Q( ∆p ) = QN , (3)
proportional pressure reducing valve. This type of valve is ∆pN
intended for limiting actuator pressures. The limited output
pressure is proportional to the control current in the pilot where QN and ∆pN are the nominal flow and nominal
section of the valve. A hydraulic scheme of the valve is pressure drop across the orifice. Respectively, the flow in the
shown in Fig. 4. laminar regime (0 ≤ ∆p ≤ ∆ptr) is given by
The main spool of the valve has two operations, reducing ∆p  ∆p .
and relieving. In the reducing operation the valve allows flow Q(∆p) = QN 3−  (4)
2 ∆pN ∆ptr  ∆ptr 
from supply inlet until the output reaches the pressure
determined by the pilot section. If the output pressure The chain of orifices and fluid fields in a hydraulic system
exceeds the pilot pressure, the output is relieved to the tank model may be relatively long and complex, since even a
pressure. Due to the reduce/relieve construction, a typical single hydraulic component may contain several fluid fields
fault for this type of valve is instability, which causes and orifices. To increase the computational efficiency of the
vibration to system pressure levels and disturbs the model, it may be worthwhile to simplify the model by
functionality of the actuators. combining some of the orifices and fluid fields. Usually this
The fault is detected measuring pressure levels from the does not affect the overall modelling accuracy significantly.
system. The waveform of pressure vibration generated by the
faulty valve is not perfectly sinusoidal, but it appears roughly
around the valve natural frequency. In order to have time
domain fault residuals that represent certain frequencies of a ACTUATOR
signal, the fault residuals are generated using the discrete Q p
wavelet transform.
The discrete wavelet transform is performed to the FLUID
pressure measurement using symlet-wavelet [6], and the FIELD
wavelet coefficients that represent the frequencies near the
valve natural frequency are chosen for fault residuals. The p Q
time-domain presentation of the fault residuals is generated
ORIFICE
calculating the floating root-mean-square (RMS) values of
the chosen wavelet coefficients. Q p

C. Simulation model FLUID

FIELD
Practically all hydraulic components can be modelled as a
combination of fluid field pressure and orifice flow equations p Q
as shown in Fig. 5. For a fluid volume V the derivative of PUMP
pressure p is given by
dp Beff  ∂V 
dt
=
V 
∑ Q − ∂t  , (1)
Fig. 5. Modelling hydraulic systems.
where ΣQ is the net flow into V. In (1) it is assumed that the
effective bulk modulus Beff of the fluid volume is constant
Fig. 6. Simulated test input and fault residuals with a healthy valve. Fig. 7. Simulated test input and fault residuals with a faulty valve.

The simulation model of proportional pressure reducing

valve consists of pilot section and main spool. The purpose
of the pilot section is to generate a control pressure that is
needed to control the position of the main spool.
Faults are generated by adjusting model parameters. The
valve model will become unstable if the delay of control
pressure in the pilot section is increased. This can be done by
increasing the friction in the pilot spool.

V. RESULTS

A. Simulation

Simulations were conducted in the case of no fault and a

single failed pressure valve. In practice, multiple faults of
different levels of severity may occur at the same time.
However, this is not a problem, since the test sequence and
pressure sensor location can be chosen in a way that other Fig. 8. Test input and fault residuals, measured from a real system
failed components do not cause false fault detection for a containing a faulty component.
healthy component.
Simulation results in the normal condition without faults
are shown in Fig. 6. A healthy pressure reducing valve B. Experimental test
reaches quickly a new steady pressure level. The fault
residuals show only a small peak in the transient state. After An experimental test was arranged using a real harvester.
the valve has gained a steady pressure level, the residuals The component to be tested was chosen to be an old, partially
return to the idle level. The idle level in the residuals unstable valve, which was still usable. The instability of the
contains background noise, which is caused by the supply valve was still so insignificant, that the fault would not have
pump and other pressure valves. been noticed by the machine operator in the normal working
The simulation results for a faulty component are shown in conditions. However, it was expected that the component
Fig. 7. After the stepwise change in the input is made, the would fail in the near future.
faulty valve is unable to regain a stable state, which causes The same test sequence that was used in the simulations
vibration in the system pressure levels. The fault residual was also used also in the experimental tests. The results
values rise considerably and the fault is easily detected. which are very much similar to the simulation results are
It is expected from the simulations that a proportional shown in Fig. 8. After the stepwise change is made in the
pressure reducing valve, with impending instability, starts to valve input, the faulty valve is unable to regain a stable state
lose its ability to regain a steady pressure level after a as expected, but the amplitude of the vibration is smaller than
stepwise change in the pressure setting. A completely faulty in the simulations. However, the residuals are significantly
valve is unable to reach a steady state at all. higher than in the idle state and the fault can easily be
detected.
C. Discussion was developed and the discrete wavelet transform, a
normally operating hydraulic valve can be distinguished from
The simulation and experimental results support each a faulty one. Impending faults are detected and can be fixed
other, though the fault residual values were slightly lower in before they cause degradation in the performance of the
the experimental test than in the simulations. The difference harvester.
was likely due to the idealisations of the model and the fact
that the valve used in the experimental tests was not totally VII. ACKNOWLEDGMENT
failed.
The test sequence used in this case was very simple The authors gratefully acknowledge the funding received
containing only open-loop control and stepwise changes. If from Sim-Serv for writing and presenting this paper.
needed, the RCP procedure allows the design of more
complicated test sequences with closed-loop control and VIII. REFERENCES
more sophisticated algorithms.
[1] Örn, J., Timber harvesting and long-distance timber
VI. CONCLUSIONS transportation costs in 2002, Metsäteho, 2003.
[2] Sirén, M., One-grip harvester operation, its
Rapid Control Prototyping reduces significantly the time silvicultural result and possibilities to predict tree
and costs of implementing a prototype of new function. damage (in Finnish), Finnish Forest Research Institute,
Although thorough modelling of the process may take a lot Research papers 694, 1998. 179 p.
of effort, the advantages gained afterwards are numerous. [3] dSPACE GmbH, Solutions for Control, Catalog 2003.
Simulation study combined with automatic code generation [4] dSPACE GmbH, datasheet of MicroAutoBox.
of the prototype controller eases the design process and [5] Chiang, L., Russel, E., Braatz, R., Fault Detection and
reduces significantly the total engineering work in the Diagnosis in Industrial Systems, Springer, 2001.
prototype testing phase. [6] Daubechies, I., Ten lectures on wavelets, SIAM, 1992.
In this test case a test sequence was designed first using [7] Ellman, A., Piché, R., “A Modified Orifice Flow
simulation and then in the real-world process. The simulated Formula for Numerical Simulation,” 1996 ASME
and measured tests produced similar results, implying that International Mechanical Engineering Congress and
the model that was used represents well the process and that Exposition, Nov. 17-22, 1996.
it can be reused in similar cases. Using the test sequence that

View publication stats