Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 49, 4, pp. 971-986, Warsaw 2011 ANALYTICAL APPROACH TO ESTIMATE AMPLITUDE OF STICK-SLIP OSCILLATIONS Jamil Abdo Sultan Qaboos University, Mechanical and Industerial Engineering Department, Oman e-mail: jdabdo@squ.edu.om Ahmed A. Abouelsoud Cairo University, Electrical Engineering Department, Egypt The objective of the presentwork is to evaluate alternative approximate techniques to determine the amplitudes of the limit cycles that evolve from stick-slip vibrations based on a mass-on-moving-belt model. The control of self-excited systems is a very interesting problem because of friction-induced self sustained oscillations which result in a very robust limit cycle that characterizes stick-slip motion. This motion should be avoided because it creates unwanted noise, diminishes accuracy and in- creases wear. The stick-slip motion produced by a mass-spring-damper on amoving belt is analyzed using the Liapunov secondmethod, which is based on constructing a positive definite function and checking the condition for which its time derivative is negative semi-definite. From this condition, an estimate of the amplitude of the velocity of the limit cycle of stick-slip motion is obtained. This estimate is found to be the zero of a certain function derived from the Coulomb friction model. An estimate of the amplitude of the displacement is also found. It is shown that the simulation results of the amplitude and the estimated amplitude are indistinguishable. Key words: stick-slip oscillations, limit cycles, Liapunov secondmethod 1. Introduction This work presents a technique that will allow useful estimation of the am- plitude of displacement and velocity of the limit cycle of stick-slip motion produced by a mass-spring-damper on a moving belt and analyzed by using theLiapunov secondmethod.Often in engineering practice, there is a need for minimization ormaximization of the friction force like in rolling and breaking 972 J. Abdo Ahmed, A.A. Abouelsoud processes. The problem of control of self-excited systems is very difficult be- cause of friction-induced vibrations. Self-excited vibrations can be frequently noticed in everyday situations, not only in the engineering practice. Noise and wear appear to be their undesirable andavoided results.Modern systems requ- ire a veryhighoperatingprecision asnecessary forworking, namely, theproper operation of various types of manipulators in modern automatic control sys- tem requires a very high operating precision. Hence, there is a strapping need to reduce the amplitude of such vibrations. This problem can be solved by ad- ditionally influenced external harmonic excitation for the amplitude vibration minimization. In some cases, it is impossible to use absorbers and the emerging stick-slip vibration is unavoidable. It is characterized by a displacement-time evolution which clearly defines stick and slip phases in which the two surfaces in contact respectively slip over each other. Themotion is governed by a static friction force in the stick phase and a velocity dependent kinetic friction force in the slip phase. The presentwork outlines a study of the estimate of the am- plitude of velocity of the limit cycle of stick-slip motion, and also simulation results of the amplitude and the estimated amplitude are compared. 2. Brief overview of the state of the art In oscillatory motion both phenomena, i.e. stick and slip, take place succes- sively, resulting in a stick-slip mode. Since friction characteristics consist of two quantitatively different parts with non-smooth transition, the resulting motion also shows non-smooth behavior. Thus, stick-slip systems belong to the class of non-smooth systems, where discontinuities occur on a surface in the state space (Utkin, 1978). Numerousmechanical interfaces are characteri- zed by some form of dry friction where the force-velocity curve has a negative slope at low velocities (Popp and Rudolph, 2004). The stick-slip vibrations are well known in many kinds of engineering systems and everyday life, e.g. like sounds formwhen a violin is played, squeaking chalks and shoes, creaking doors, squealing tramways, chattering machine tools, drillstrings, car steering systems, grating brakes and various other systems. For ease of the setup and interpretation, an idealized physical system consisting of a mass sliding on a moving belt has been considered very often. For self-excited friction induced oscillations, essentially four different instability mechanisms have been descri- bed in literature. First, the friction coefficient decreasing with relative sliding velocitymay lead to negative damping and consequently to oscillatory instabi- lity of the steady sliding state. Second,mode-coupling (sometimes also referred Analytical approach to estimate amplitude... 973 to as binary flutter or displacement dependent friction force instability) may destabilize the steady sliding state also for constant friction coefficients. Third, sprag slip, and fourth the follower force nature of the friction force have been identified as fundamental mechanisms for friction self-excited vibrations. All of thesemechanisms are amply described in literature (Spurr, 1961; Popp and Stelter, 1990; Ibrahim, 1994;Wallaschek et al., 1999; Gaul and Nitsche, 2001; Gasparetto, 2001; Hoffmann et al., 2002; Hoffmann andGaul, 2003; Abdo and Al-Yahmadi, 2009; Abdo et al., 2010), a further discussion is therefore not given here. Also when it comes to the system nonlinearities, a lot of work has already been conducted. Mainly, there are four differentmodels of stick-slip friction:Rough surfaces or surface topology model (McClelland, 1989; Meyer et al., 1998), distance- dependent or creep model (Sampson et al., 1943), velocity dependent friction model (Carlson and Langer, 1989; Nasuno et al., 1997), and phase transition model (Thompson and Robbins, 1990; Robbins and Thompson, 1991). An approximate analysis of the stick-slip vibration amplitude is conducted by Thomsen andFidlin (2003). The analysis is based on dividing themotion into two phases. The stick phase in which the velocity of motion is constant and the slip phase in which motion is approximated as circular (pure sinusoidal) motionwith a constant amplitude.The solutions of the twophases are batched together. The exact solution for this model is given in Wensrich (2006) for a simple Coulomb frictionmodel consisting of a sign function and a linear term in the relative velocity with no cubic term. One sided solution (i.e. when the relative velocity is negative) is only considered. This is also done by Thomsen and Fidlin (2003). Same model with friction is considered by McMilan (1997). The model treats velocity and acceleartion in a discontiuousmanner. The experimentally observed phenomenon is explained and aproximated by hystersis. The model is numerically integrated and the results are compared with the experimental ones to validate the friction model. Leine et al. (1998) proposed a numerical methodknownas the shootingmethod for calculation of theperiodofvibration of a one degree of freedom mechanical model. Two degrees of freedom for stick-slip motion is studied in Awrejcewicz and Olejnik (2007) in which the mass-spring- damper on a belt is used with allowable vertical motion. In this paper, stick-slip vibration is analyzed for one degree of freedom depending on the model of Thomsen and Fidlin (2003). However, a novel approach for estimating the amplitude of vibration is introduced.Aprediction of oscillations is done using theLiapunov secondmethod (Khalil, 1992), which is usually used to check the stability of equlibriumpoints.TheLiapunovdirect 974 J. Abdo Ahmed, A.A. Abouelsoud method, also known as the Liapunov second method, represents an approach to the problem of stability of dynamical systems not requiring the solution to the differential equations of motion. The condition for the derivative of a negative definite or semi-definite Liapunov function is examined from which the amplitude of the velocity of the limit cycle is estimated.The frictionmodel of Thomsen and Fidlin (2003) is used here. It is shown that the amplitude of the velocity is a zero function obatined from the frictionmodel. The estimated amplitude is thus related to the model parameters allowing for controlling the amplitude of vibration by adjusting these control parameters. Simulation results show matching between the predicted amplitude and the simulation results. 3. Analysis using the Liapunov direct method The equation of motion of a mass-spring-damper on a moving belt, shown in Fig.1, is given by MẌ+CẊ+KX =Fs (3.1) where X is the displacement, M is the mass, C is the viscous damping, K is the spring constant, and Fs is the friction force, where the dot indicates derivationw.r.t. t. Figure 2 shows a typical friction force as given inThomsen and Fidlin (2003), Ibrahim (1992), Navaro-Lopezz et al., 2004. Fig. 1. Mass-spring-damper on amoving belt Let Vr = Ẋ−V be the relative velocity of the mass, for |CẊ+KX| ¬ ¬µsMg (g is the gravity vector) and Vr =0, the pulling force is smaller than the friction force and there is nomotion (stick phase), hence MẌ =0 for |CẊ+KX| ¬µsMg and Vr =0 (3.2) otherwise Fs =−µsMg sgn(Ẋ−V )+K1(Ẋ−V )−K3(Ẋ−V )3 (3.3) Analytical approach to estimate amplitude... 975 Fig. 2. Friction function as given by (3.2) where sgn(s)=    1 for s> 0 0 for s=0 −1 for s< 0 and the equation of motion is MẌ+CẊ+KX =−µsMgsgn(Ẋ−V )+K1(Ẋ−V )−K3(Ẋ−V )3 (3.4) with µs =0.4 µm =0.25 vm =0.5 K1 = 3(µs−µm) 2vm K3 = µs−µm 2v3m Transformation of the equation ofmotion into a dimensionless form is done in Thomsen and Fidlin (2003). Let L= Mg K ω0 = √ K M t0 = 1 ω0 k1 = K1 Mω0 k3 = K3ω0L 2 M 2β= C√ KM v= Vt0 L τ = t t0 x= X L vr = VrL t0 equations of motion (3.2), (3.4) read ẍ=0 for ẋ= v and −µs ¬ 2βv+x¬µs (3.5) ẍ+2βẋ+x=−µs sgn(ẋ−v)+k1(ẋ−v)−k3(ẋ−v)3 for ẋ 6= v where the dot indicates derivation w.r.t. τ. 976 J. Abdo Ahmed, A.A. Abouelsoud Analysis of motion is carried out using the Liapunov secondmethod (Ba- nakov andGubanov, 1965; Leine et al., 1998), by splitting it into two phases, the stick and slip phase. Partition the state space R2 into three regions Ω= [−µs−2βv,µs−2βv]×{v} Ω− =R×{ẋ < v}∪ [−µs−2βv,−∞]×{v} Ω+ =R×{ẋ > v}∪ [∞,µs−2βv]×{v} In Ω, motion is linear x(τ)=x(0)+vτ (3.6) In Ω−, motion is described by ẍ+2βẋ+x=µs+k1(ẋ−v)−k3(ẋ−v)3 (3.7) and in Ω+, motion is described by ẍ+2βẋ+x=−µs+k1(ẋ−v)−k3(ẋ−v)3 (3.8) We consider two cases: Case 1: Themotion starts inΩ Case 2: Themotion starts in Ω− or Ω+. Case 1: Assume that the motion starts in Ω, we show that the trajectory described by (3.6) and (3.7) returns to Ω. Thus the solution maps into itself. Since the exit point from Ω is unique (x = x0, ẋ = v), (xω0 = µs − 2βv) this is a unique trajectory, call it the stick-slip limit cycle. However to be a true limit cycle it should have the property that any trajectory starting in its neighborhood must converge to (or diverge from) it Leine et al. (1998). This stick-slip limit cycle is the union of two motions; one given in (3.6) and the other being the solution to (3.7). To analyze the solution to Eq. (3.7) starting from (x0,v), using the Lyapunov method, a change of variables is conducted to move the equilibrium point to the origin. The equilibrium displacement x=µs−k1v+k3v3. Let y=x−x, hence ẏ= ẋ, and ÿ+(2β−k1+3k3v2)ẏ−3k3vẏ2+k3ẏ3+y=0 (3.9) Let h(ẏ)= (2β−k1+3k3v2)ẏ−3k3vẏ2+k3ẏ3 (3.10) then ÿ+h(ẏ)+y=0 (3.11) Analytical approach to estimate amplitude... 977 Make the following change of variables: ẏ= z, hence ż= ÿ=−y−h(z) z̈+h′(z)ż+z=0 (3.12) where h′(z) is the derivative of hw.r.t z. Let x1 = z, x2 = ż+h(z), then ẋ1 =x2−h(x1) ẋ2 =−x1 (3.13) Note that x1 = ẏ, x2 =−y. Consider the Liapunov function W = 1 2 (x21+x 2 2) (3.14) Time differentiating W along (3.13), we obtain Ẇ =−x1h(x1) (3.15) Figure 3 shows the function h(x1). It has three roots r2 = 3 2 v+ √ D > 0, r1 = 3 2 v− √ D< 0 and r0 =0, if the belt velocity v is less than vmax = √ k1−2β 3k3 (3.16) where D= (3v 2 )2 − 2β−k1+3k3v3 k3 (3.17) Fig. 3. The polynomial h(x1) and its approximations in D1 and D2 The range of h(x1) is divided into three intervals x1 > r2, r2 ­ x1 ­ r1 and x1 x1ĥ2(x1) in D2 and Ẇ ¬−x1ĥ2(x1)¬ 0 (3.20) Since 0.5r2 = x21 +x 2 2 = ẏ 2 + y2 it follows that in D2: dr 2/dt ¬ −x1ĥ1(x1) or r2 ¬ − ∫ x1ĥ1(x1)dt and the trajectory of motion is contained insi- de the trajectory described by DE: ÿ + ĥ1(ẏ) + y = 0 starting at (y0,v), (y0 =x0−x)∈D1. In D2, dr 2/dt ¬ −x1ĥ2(x1) or r2 ¬ − ∫ x1ĥ2(x1)dt and the tra- jectory of motion is contained inside the trajectory described by DE: ÿ+ ĥ2(ẏ)+y=0∈D2. We choose ĥ1(x1) = −a1x1, a1 > 0 in D1 and ĥ2(x1) = −a2(x21 − r21), a2 > 0 in D2. We analyze the trajectory of motion in the two regions D1,D2. In D1 ÿ−2a1ẏ+y=0 (3.21) which is a linear DE and has two complex poles in the right half plane at a1± j for a1 ≪ 1. The solution to which is y(t)= ea1t(y0cost+(v−a1y0)sin t) (3.22) If ẏ= r1, y= y1 ≈ √ y20 +(v−a1y0)2−r21 approximately for a1 ≪ 1. In D2 ÿ−a2(ẏ2−r21)+y=0 (3.23) or dẏ2) dy −2a2(ẏ2−r21)+2y=0 (3.24) Analytical approach to estimate amplitude... 979 This is a first order linear differential equation ẏ2 which has the solution ẏ2 =e2a2(y−y1) ( ẏ21 − 2a2y1+2a 2 2r 2 1 +1 2a22 ) + 2a2y+2a 2 2r 2 1 +1 2a22 (3.25) starting from (y1, ẏ1). If ẏ1 = r1 then the trajectory will move clockwise until ẏ= r1 again and y reaches the value y2 which satisfies the equation (1+2a2y2)e −2a2y2 =(1+2a2y1)e −2a2y1 = c (3.26) For a solution to exist c < 1. Solving for y2 and y1, we get using the Padé approximation of e−x =(1+x+x2/2)−1, a2y2 =(1−c− √ 1−c2)/c< 0 and a2y1 =(1− c+ √ 1− c2)/c> 0. Eliminating c, we get a2y2+a2y1 =2 1− c c a22y2y1 =−2 1− c c =−a2y1−a2y2 Thus y2 = −y1 1+a2y1 (3.27) Hence, the absolute value of y2 is smaller than y1. If y=0, ẏ= ẏm ẏ2m =e −2a2y1 ( ẏ21 − 2a2y1+2a 2 2r 2 1 +1 2a22 ) + 2a22r 2 1 +1 2a22 (3.28) When reaching (y2,r1), the trajectory enters D1 again and has an equation y(t)= ea1t(y2cost+(r1−a1y2)sint) (3.29) which hits the line y2 = v at approximately y= y3 y3 ≈− √ y22 +(r1−a1y2)2−v2 (3.30) atwhich themotion enters the stick phase again since y3 ­−y0. The smallest slip velocity is greater than r1+ ẏm and the amplitude of the displeacement is smaller than √ y20 +(v−a1y0)2. This completes the analysis of Case1. Case 2: If the trajectory of motion starts in Ω−, then it is described by Eq. (3.7). Shifting the equilibrium again, we obtain Eq. (3.9) which possesses a stable limit cycle in the whole plane (Leine et al., 1998) (provided that there is no other type of motion). Any trajectory starting in the plane would converge to this limit cycle, call it the right limit cycle. 980 J. Abdo Ahmed, A.A. Abouelsoud Similarly, if the motion starts in Ω+ the equation of motion is (3.8) with the equilibrium point x = −µs−k1v+k3v3. Shifting this equilibrium point to the origin, we obatin the same Eq. (3.9) which also possesses a stable limit cycle in thewhole plane (provided that there is no other type ofmotion). Any trajectory starting in the plane would converge to this limit cycle, call it the left limit cycle. The distance between the two origins is 2µs, which equals the length of Ω. If the right and left limit cycles intersect the set Ω, then a trajectory starting at any point in the plane must enter the set Ω in trying to reach one of the limit cycles. Figure 4 shows these limit cycles and their intersections with the set Ω. Thus, eventually every trajectory will enter the set Ω, and hence reach the slip-stick limit cycle. This completes the analysis of Case 2. Fig. 4. The left and right limit cycles and there intersection with Ω 4. Simulation results The dynamic model in Eq. (3.7) is simulated using the parameters used in Thomsen and Fidlin (2003), where β = 0.05, v = 0.25, µs = 0.4, µm = 0.25, vm = 0.5, k1 = 0.45, k3 = 0.6. For the selected parameters, the zeros of h(ẏ) are 0, 1.1074,−0.3574, the negative zero is an estimate of the amplitude of velocity. Figure 5 shows the stick-slip limit cycle starting from (0,0.25). Figure 6 shows both the displacement and velocity, where the stick phase is obvious. Figures 7 and 8 show the the trajectories with stating points (0,0.125) and (0,0.5), respectively. Both converge to the slip-stick limit cycle. Note that the lower slip velocity predicted in Thomsen and Fidlin (2003) is equal to A = 0.7265, which is about two times the absolute value of our Analytical approach to estimate amplitude... 981 Fig. 5. Stick-slip limit cycle with the initial point (0,0.25) Fig. 6. Displacement and velocity with the initial values (0,0.25) Fig. 7. Trajectory with the initial point (0,0.125) converges to the slip-stick limit cycle 982 J. Abdo Ahmed, A.A. Abouelsoud Fig. 8. Trajectorywith the initial point (0,0.5) converges to the slip-stick limit cycle Fig. 9. Trajectory with the initial point (0,0.45) converges to the slip-stick limit cycle with belt velocity 0.45 Fig. 10. Displacement and velocity with belt velocity 0.45 (initial point (0,0.45)) Analytical approach to estimate amplitude... 983 estimated value. This shows that our estimated amplitude is more accurate than theonepresentedbyThomsenandFidlin (2003). Thedisplacement range in the existed work is −0.45