MountainBikeSuspensions Design-Opt
MountainBikeSuspensions Design-Opt
design optimization
Craig Good and John McPhee
Systems Design and Mechanical Engineering, Unive rsity of \ Vaterloo, Canada
Abstract
Using a dynamic model of a rear-suspension mountain bicycle, numerical methods are
emp loyed to optimize the geometric variables associated with the suspension design. In
particular, a genetic algorithm is used to minimi ze the maximum value of the frame's
rotation, in order to minimize the chain-suspension interactions felt by a rider. Two sets
of constraint equations are included in the optimization problem. The first set
constrains the wheelbase to be constant for all designs , thereby ensuring equivalent
hand ling properties between designs. The second set is required to avoid infeasible
designs associated with a co ll apsed suspension. Resu lts using four geometric parameters
as design variables are in good agreement with those obtained using a more complex
dynamic model, and with those obtained experimentally. Our optimal design is
manufacturable, since the locations of pivots and suspension members are such that no
interference will occur between moving parts .
Keywords: design optimi zation, genetic algorithm, mo untain bicycle, rear suspe nsion
Genetic algorithm
with best fitn ess. T he design associ ated with thi s In additi o n, by o nly co nsiderin g th e spring-
indi vidu al is taken as th e so luti o n to th e o ptimi- damper, o ne has no id ea ho w changes in th e design
za tion pro bl em. variabl es affect the actu al moti o n o f the rid er.
The geneti c al gorithm is well suited fo r o pti- A susp ensio n th at minimi zes the energy absorbed
mi zing multibody systems. A G A does no t require by th e sprin g-d amper m ay no t minimi ze th e degree
gradi ent in fo rm ati o n fo r th e o bjecti ve fun cti on o r o f chain-s uspension interactio ns th e ri der actu all y
its con straints. Thi s ca n be important sin ce th e fee ls.
o bj ective fun cti o n in a multibody sys tem often C o nsid ering these facto rs, an altern ative o bj ec-
invo lves th e computati o nally exp ensive procedure ti ve fun ctio n f was pro posed and used for th e
o f integratin g th e equati o ns o f m o ti o n. Also, th e o ptimi za ti o n of th e rea r suspensio n:
probabili ty o f findin g des ig ns at o r near the glo bal
o ptimum with a geneti c algo rithm is hi g her th an f = [m ax03(t)- min 03( t)JI;~ + '+'n (1)
fo r mos t no nlin ea r p rogrammin g techniqu es (Aro ra where 03 is th e angle of rotati o n of the frame and
et al. 199 5). rid er, des ig nated as Body 3 in th e fo ur- body bi cycl e
Th ere are, however, a few di sa dva ntages to usin g model shown in F ig. 2. By using fas an o bj ecti ve
GAs for thi s type o f pro bl em. Sin ce th e algorithm is fun cti o n , o ne seeks to minimi ze th e oscill ati o n
probabili sti c, th ere is no rea l proof th at th e des ig ns amplitud e o f 0 3 durin g stea dy-state pedalling,
found converge to the o ptimum . In ad diti o n , a th ereby minimi zin g th e pitchin g motio n of th e
mu ch larger number o f o bj ecti ve fun cti o n eva lu- fram e and rid er.
ati o ns are required than for o th er o ptimi za ti o n To ca lcul ate the fitn ess fun cti o n va lu e, the
m eth ods. equati o ns o f m otio n fo r th e fo ur-body bi cycl e were
Th e description o f th e techniqu es used in a integrated usin g a specifi c set o f des ign va riabl es.
geneti c algo ri thm prese nted here is intend ed as an T he minimum and maximum va lu es fo r 03 we re
ove rvi ew. Fo r more detail s th e reader is enco uraged determin ed fro m th e coordin ate res po nse over
to consult Srini vas & P atnaik (1994), Go ldb erg th e tim e interva l t i ::::; t ::::; t 1 , where t i and t 1 arc
(1989) and Baum al (1997). ca refull y selected to ensure th at th e va lu es extra cted
Baumal impl em ented th e simpl e ge neti c algo - are fro m th e stea dy-state regime. T he difference
rithm , fo rm ali zed by Go ldberg, in a so ftware pack- betwee n th e maximum and minimum va lu e o f 03
age call ed 1VlEc 1IAGEN, ava il abl e fro m th e Moti o n represents the amplitud e of oscill ati o n o f the
Research G roup at th e U ni versity ofWaterl oo . T hi s coo rdin ate over th e spec ifi ed interva l.
packa ge was used to o ptimi ze th e geo m etry o f
th e rear suspensio n for th e fo ur-body bike m odel.
Fitness function
Design optimization
Rear Suspension
Body 2
4 Variable Ana lysis
'y/ody3 Frame & Rider
4 Variable Analysis
Design Space · Node h
This corresponds to the loca l pos1t1on of nod e h. Tab le 1 Bounds on design l'ariab lcs
Discussion of results
From th e simulation of the original fou r-body
bi cycle model given in Good & McPhee (1999), the
va lue of the fitness function was ca lcul ated to be
10.3 mrad o r 0.6°.
By runnin g the opti mi zatio n process, the fi tness
functi o n wa s substantiall y red uced. In the four
va riabl e ana lys is, three of the fo ur runs reduced
Optimized Des ign
the fitn ess to less than 0.01 mn1d . This 1s a
Figure 5 Optimi zed s u s pen~ion. sig nificant improve m ent (red ucti o n o f 99%) ove r
Tab le 2 Optimi zation res ults was located 11.6 cm above the bottom bracket and
2.7 cm reanvards when minimizing the pitching
Variable Run 1 Run 2 Run 3 Run 4
motion of the rider. Considering the differences in
Px (m) 0.0667 0040 0.0667 - 0.0933
modelling and differences in the objective func-
Py (m) 0.220 0.200 0.280 0.120
qx (m) - 0.414 - 0.414 - 0.307 - 0.414 tions, this is reasonable agreement. In future work,
qy (m ) - 0.700 - 0.700 - 0.725 - 0.700 the clamper energy could be used as the fitness
CPU time (s) 716 861 751 986
function, to allow direct comparison of the results
Fitness evaluations 180 253 237 313
Generations 14 14 12 26 with those of Wang and Hull.
Minimum fitness (mrad) 0.00* 0.00* 0.06 0.00* It is important to consider that the results
* M echagen cannot output res ults beyond 5 decimal pla ces presented here are specific to the bike topology
(in rad). and geometry as given in Good & McPhee (1999).
In the bicycle market, there are many different
suspension designs available in many different si zes.
the original design. Runs 1, 2 and 4 converge to an The techniques developed here are general and can
optimal solution where 1x = -0.414 m and fj y = be extended to the design optimization of any type
0.700 m. In contrast, the results found for Px ~nd of suspension of any size.
Py do not correspond to specific optimum values, Finally, we note that our optimal design is
but instead fit into a narrow band of possible manufacturable. The locations of the pivots and
values. From this, we concluded that a design with suspension members are such that no interference
minimum fitness is not as sensitive to the variables will occur between parts. The pivots and springs
Px and Py as it is to 1x and 1T The optimal pivot can be successfully located in the proper positions
point corresponds to a point which is located without getting in the way of other moving parts.
1 1.6 cm above the bottom bracket and 2. 7 cm
behind the seat tube as shown in Fig. 5.
Conclusions
It is difficult to explain with certainty why
particular design variables are found by the opti- Using a four-body dynamic model of a mountain
mization process, due to the probabilistic nature of bicycle with a rear suspension, we have employed
the genetic algorithm. Run 3 found a much numerical methods to optimize the geometric
different solution whose fitness is less optimal than variables associated with the suspension design. In
runs 1, 2 or 4. Convergence of the algorithm is not particular, we have used a genetic algorithm to
a proof that the result is an optimum. In future minimize the maximum value of the frame's rota-
work, this result could be used as a starting point tion , which we equate to minimizing the chain-
for a SQP (Sequential Quadratic Programming) suspension interactions felt by a rider. This was
opt1m1 zation algorithm, which will converge accomplished while satisfying constraints that
quickly to a local optimum. ensured a constant wheelbase for all designs (to
Data from the optimization is directly compar- ensure equivalent handling properties between
able to optimi zation results found by Wang & Hull designs) and avoided infeasible designs associated
(1997) and Needle & Hull (1997). Wang and Hull with a collapsed suspension. Results using four
found the optimum pivot point to lie 10-12 cm geometric parameters as design variables were in
above the bottom bracket, and Needle and Hull good agreement with those obtained using a more
found the optimum pivot point to be 8.4 cm above complex dynamic model (Wang & Hull 1997) and
the bottom bracket. Both attempts minimized with those obtained experimentally (Needle & Hull
energy lost in the spring-clamper. It is important 1997), even though we have used a significantly
to note that these previous attempts did not allow different objective function.
the pivot to move fore and aft of the seat tube. It was found that the correct placement of the
Using our optimization criteria , the optimal pivot pivot point, with respect to the frame, was the
most signifi ca nt geometric suspension variab le behaviour. :\1aste r's Thesis, U ni ve rsity of \ \laterloo,
that cou ld be used to reduce chain-suspension \ \'aterloo, Ontario, Canada.
interactions. These interactions we re not as sen- Go ldberg, D .E . ( 1989) Genetic Algorith711s in Search,
Opti711i:,ation and J\!lachi11e Leaming. Add iso n-\ Vesley
sitive to other geometri c suspensio n parameters.
Publishing Compan)·, Readi ng, .VIA, USA.
This agrees with earli er results found by \!Van g Good , C. & NlcPhee, J. (1999) D ynami cs of mou ntain
and Hull. bi cycles with rear suspensions: mode lling and simul ati o n.
Spom Engineering, 2, 129- 1-J.3.
H o lland , ].H. ( 197 5) Adaptation in .Vatum/ rmd r l 11ificial
Acknowledgements
Syste711s. U niversity of Michigan Press. Ann Arb o r, :\U,
Financia l support of thi s resea rch by th e Natura l U.S.A.
l\'eedl e, S.A. & Hull , i\'l.L. ( 1997) An off- road bicycle with
Sciences and E ngin eerin g Resea rch Council o f
ad justa bl e suspensio n kinematics. J ouma/ of ,Hechm1iwl
Canada is gratefu ll y acknowledged.
Des1~~'11, 119, 3 70-3 75.
Srini vas, NI. & Patnaik, L. .Vl. ( 199-J.) Ge netic algo rithms:
a survey. Co111p11ter, June, 17-26.
References
\ Vang, E.L. & ll ull, ,\LL . ( 1996) ,.\ model for determining
Arora, ].S. , E lwakeil, 0.A. & Chahandc, A.l. ( 1995) Globa l rider induced energy losses in bicycle suspe nsion
o ptimi za ti o n methods for engin ee rin g app li cations: systems. Vehide Syste711 Dy11a711ics, 25 , 223 -2 -J.6.
a review. Srmctuml Opti711i:,ation, 9, 137- 15 9. \Vang, E.L. & IIull , Nl.L. ( 1997) 1Vlinimi za ti o n of
Bau mal , A.E. ( 1997) Automated design of mechan ica l peda ling indu ced energy losses in off-road bi cycle rea r
sys tems through num eri ca l o ptimi za tion of dynam ic suspe nsio n systems. Vehide System Dy11m11ics, 28, 29 1-3 06.