Proceedings of the Twenty Second Nordic seminar on Computational Mechanics, Aalborg

University, ISSN 1901-7278 DCE Technical Memorandum No. 11, (2009)

An Adams/View-Matlab Computational Interface

for Clustered Optimization of Washing Machines

Thomas Nygårds and Viktor Berbyuk

Department of Applied Mechanics
Chalmers University of Technology, Göteborg, Sweden
e−mail: [email protected], [email protected]

Summary This paper describes an Adams/View-Matlab environment for parallel/clustered calculations

running on ordinary workstations. The system was developed for sensitivity analysis and bi-objective
optimization of washing machine performance by using a dynamic model built in the commercial multi-
body software MSC.Software Adams\View. Together with statistics of performance of the system, results of
a bi-objective optimization of a selection of structural parameters are presented.
Keywords: Bi-objective optimization, parallel computing, washing machine dynamics, vibration control

Introduction
Optimization of dynamical systems is often computationally demanding. Not seldom days or
weeks of computational time have to be spent on solving an optimization problem. Several things
influence on the CPU-time it takes to solve a specific problem. Always important are the number
of parameters to be optimized and the complexity of the model which directly affect the time it
takes to solve its equations. The number of conditions for which the model must be evaluated has
also a strong influence on CPU-time. For instance, in the case of optimization of a car suspension
it could be different road types or handling situations that the suspension needs to perform in. The
amount of time it takes to prepare a model for simulation has also an effect on the time
consumption. Sometimes heavy in data is needed to be loaded from disk before running.
The paper aims to present a developed computational environment for clustered bi-objective
optimization of washing machine dynamics on a set of its operational conditions.

The washing machine model

A washing machine is an object familiar to almost everyone. Many people associate washing
machines with noise and vibration, and locate their machine to remote parts of their house or
apartment. With extreme cases of unbalanced load, which is the main reason to vibrations and
noise, together with a fail of the unbalance detection electronics a washing machine behaviour
called “walking” can occur. Walking happens when the normal forces of the floor become too low
in relation to the lateral forces and the foot loses its grip.
To deal with these and other issues related to vibration analysis, control and dynamics
optimization, a model of a modern in-production washing machine has been developed. Work has
been performed during the last couple of years in collaboration with the washing machine
manufacturer Asko Appliances AB [1, 2].
The work has resulted in a rigid multi-body model which has been implemented in the
commercial software Adams/View. It can be seen together with relevant inner structural
components of a physical machine in figure 1. Model parts are constructed using CAD drawings
giving accurate inertia properties and joined with appropriate constraints. Remaining structural
components of importance have been measured separately with dynamic measurement test-rigs
resulting in separate sub models for dampers, springs, rubber bushings and feet. Details on the sub
models can be found in [1, 2].
Proceedings of the Twenty Second Nordic seminar on Computational Mechanics, Aalborg
University, ISSN 1901-7278 DCE Technical Memorandum No. 11, (2009)

Figure 1: Some of the inner parts of a washing machine and the Adams\View model representation.

The optimization problem

In this paper two criteria of high importance in washing machine dynamics will be used. The first
is to keep the tub free from hitting the housing during washing. The second is to limit the
propagation of forces through the structure to minimize vibration impact on surroundings and to
prevent the machine’s walking behaviour [2]. The first objective to be minimized is defined as
follows:
=
p
(
ℑK max max ( ∆X p (t ) − ∆X max
t
p= )
) , p 1,2,3...9, ∀t ∈ [0, T ] (1)

where ∆X p and ∆X max

p are movement and movement margins in three directions at the points p of
the tub which have been defined as critical by the manufacturer. The second objective to be
minimized is the sum of the RMS-values of the vertical forces Fi z , (i=1,2,3,4) at the four feet of
the machine during the simulation time T and is written as
T
4
1
( Fi z (t ) ) dt
2
ℑD =∑
i =1
∫
T 0
(2)

To be a feasible product to sell, a washing machine has to be able to handle different amounts
of load and different imbalances. To reflect these different operational conditions three critical
load cases with respective drum rotational excitation schemes are defined as follows:
1. Constant load of 1kg placed in the front of the drum whilst spinning up to 800rpm with a
gradient of 80rpm/s.
2. Constant load of 0.3kg placed in the middle of the drum whilst spinning up to maximum
spin speed.
3. Maximal load of the machine distributed evenly in the drum with an exception of 1 kg,
which is placed in the front of the machine.
The bi-objective optimization problem of washing machine vibration dynamics on a given set of
operational conditions is stated as follows.
Problem A. It’s required to determine the vector of structural parameters d* and state vector
x* (t ) which satisfy the variational equation min {F [ d, P, x(t )]} = F [ d* , P, x* (t )] subject to the
d ,x ( t )

differential equations of motion x = f ( x, t , d, P ) and constraints Bl ≤ d ≤ Bu . Here F =[ ℑK , ℑD ]

T

P = [ P1 , P2 , P3 ] where Pi is a vector of input parameters to the operational condition i, (i=1,2,3).

T
Proceedings of the Twenty Second Nordic seminar on Computational Mechanics, Aalborg
University, ISSN 1901-7278 DCE Technical Memorandum No. 11, (2009)
The computational environment and optimization results
To solve Problem A a computational environment has been created. It consists of one computer
equipped with Matlab running an optimization algorithm. To this computer an arbitrary amount of
computational processes, on the same or on other computers, running Adams/View can be
connected through a developed interface.
The dynamical system model opened in Adams\View contains a macro which when it runs
creates a unique ID for the program instance and registers it on a network path. It then loops
waiting for commands or simulation in data. The in data for the simulations are created in Matlab
by the optimizing algorithm and written to files by the developed interface which marks them with
the ID of an available node. The interface enables parallel start of simulations if the optimizing
algorithm can take advantage of this. One optimizer which works in parallel is the used MATLAB
function ‘gamultiobj’ from the ‘Genetic Algorithm and Direct Search’. This optimizer creates a set
of in data (called generation) at each optimization step and makes it available for the interface
function to start. The interface searches for available computers to start simulations on, searches
for and processes calculated files, sends status messages to the user and restarts crashed
simulations when the Adams/View program or a computer is the reason for the crash.
As an example of a solution of Problem A the following results are presented. The resulting
Pareto front visible on the left of figure 2 was obtained by the optimization algorithm after more
than 3500 evaluations of sets of parameters d.

Figure 2: Resulting Pareto front from optimization together with a selection of resulting parameter values.

Together with the Pareto front some of the simulations are plotted with color depending on which
of the points p where active in equation (1), i.e. at which point the maximum movement in relation
to the allowed occurred. On the right side of figure 2 two of the variables which values results in
optimal performance are plotted against the respective objective. The conflicting nature of the two
objectives can clearly be observed.

Cluster performance
As a performance test a special simulation case (not connected to Problem A) was created. Here all
parameters were held constant during simulation to enable predictable calculation time. Varying
the time of simulation T, resulted in the relation for the cluster efficiency showed in figure 3
Proceedings of the Twenty Second Nordic seminar on Computational Mechanics, Aalborg
University, ISSN 1901-7278 DCE Technical Memorandum No. 11, (2009)
peaking at 96% for a T=251 s simulation. The data of each point is based on an average of 100
simulations. A curve is fitted to the points following the function g(t) in equation (3)
 a 
g (t ) = 100 ×  1 − − ct  (3)
 t+b 
where a=5.45, and b=5.44 and c=0.00101 giving a goodness value R2=0.997.

Figure 3: Efficiency of the cluster as a function of effective simulation time.

So, it can be stated that the average constant delay for results file administration for each
simulation can be estimated to 5.44 seconds. The administration includes file reading and writing
in Adams/View and input file-generation in Matlab. Included in this is a random delay which can
be anything between 0 and 4 seconds depending on when for example Adams searches for files in
relation to when it is written by Matlab. The random delay can be tuned further if found necessary.
A file reading and writing time depending on the size of the output file is reasonable and
represented here by coefficient c.

Concluding remarks
An Adams/View-Matlab based interface has been developed for parallelized computation and
constrained multi-criteria optimization of washing machine dynamics on a set of critical
operational conditions.
Implementation of the developed tools on ordinary workstations has proved calculation
efficiency of up to 96% in case of optimization of the dynamics of a modern washing machine.
By using the developed interface and tools a Pareto front has been found giving the domain of
realistic structural parameters with optimal dynamic and kinematic performance.

References
[1] T. Nygårds, V. Berbyuk, A. Sahlén: Modelling and Optimization of Washing Machine Vibration
Dynamics. Proceedings of the 9th International Conference on Motion and Vibration Control (MOVIC
2008), September 15-18, 2008, Technische Universität München, Munich, Germany.
[2] T. Nygårds, Modeling and Optimization of Washing Machine Vibration Dynamics during Spinning.
Thesis for Licentiate Engineering 2009:1, Department of Applied Mechanics, CHALMERS, 2009.