“… cette résistance est à peu près égale au tiers de la pression.”
“… this resistance is roughly equal to one third of the pressure.”
— Guillaume Amontons, De la résistance causée dans les machines, Mémoires de l’Académie Royale des Sciences, 1699
Introduction to Friction
The box, pushed by \(P\).
Its free body diagram.
Figure 5.6.1: A box on a rough floor, pushed by a horizontal force \(P\). The friction force \(F_f\) acts along the contact surface, opposite to \(P\).
The box with the mass \(m\) kg is exposed to an external force \(P\) in Figure 5.6.1. Doing a free body diagram and applying Newton’s third law reveals a force \(F_f\) acting opposite to \(P\) along the contact surface. The magnitude of this friction force depends on the type of surface and the normal force \(N\). The relation between the normal force and the friction force is modelled by Coulomb’s law:
\[
\boxed{ F_f \leq \mu_s N }
\tag{5.6.1}\]
where \(\mu_s\) is known as the static coefficient of friction. Amontons stated the proportionality in 1699, and Coulomb confirmed it for dry friction in 1781 and separated the friction at rest from the friction in motion. The law assumes that the friction force grows in proportion to the normal force. That holds over a wide range of loads for most pairs of surfaces, but not for every load and every material, as later in this chapter shows.
Figure 5.6.2: Relation between the external force \(P\) and friction force \(F_f\).
As the force \(P\) exceeds the largest possible frictional force \(F_{f,\max}\) the box will start to slide and get an acceleration, thus the model transitions to what is known as kinetic friction and the opposing force is given by \(F_f=\mu_k N\), where \(\mu_k\) is the kinetic coefficient of friction. This is where statics ends and dynamics continues.
Coefficients of friction
A coefficient of friction belongs to a pair of surfaces in a given state, not to a material. Two mild steel surfaces give \(\mu_s = 0.74\) clean and dry, and a sliding coefficient of 0.09 to 0.19 with oil between them. Table 5.6.1 collects measured values for common pairs from Marks’ handbook [1], as reprinted by NASA [2]. The values are in the friction module of MechanicsKit, which the notebook below reads. Each value is one measurement under the conditions of its source, so a design that relies on friction works with a range and checks both ends of it, and a design that relies on it closely measures the coefficient for its own surfaces.
To look up a single pair, choose the two surfaces below. The second list offers only the surfaces the source has measured against the first.
Table 5.6.1: Coefficients of friction of clean surfaces, from Marks’ Standard Handbook [1] as reprinted in Table IV of [2]. Only the pairs with both dry coefficients are listed. The lubricated values span the oils tested in the source; an empty cell is a value the source does not give.
from mechanicskit import frictionfriction.table('pairs', source=False)
Surfaces
\(\mu_s\), dry
\(\mu_s\), lubricated
\(\mu_k\), dry
\(\mu_k\), lubricated
Hard steel on hard steel
0.78
0.11–0.23
0.42
0.029–0.12
Mild steel on mild steel
0.74
0.57
0.09–0.19
Cast iron on cast iron
1.10
0.15
0.064–0.07
Brass on mild steel
0.51
0.44
Copper on mild steel
0.53
0.36
0.18
Aluminium on mild steel
0.61
0.47
Aluminium on aluminium
1.05
1.40
Oak on oak, along the grain
0.62
0.48
0.067–0.164
Leather on oak
0.61
0.52
A calculation takes its coefficient from the module instead of from a number typed into the notebook; the MechanicsKit documentation shows how. The module holds three more tables for the Machine Elements part: the friction classes of bolted joints after VDI 2230 [3], the brake and clutch linings of Budynas and Nisbett [4], and the slip coefficients of interference fits after DIN 7190 [5]. friction.table('bolts'), 'linings', 'shrink' and 'pressed' print them, and friction.SOURCES names the source of each.
WarningDo not rely blindly on a table
Each value in the table was measured once, on clean surfaces prepared in a laboratory, and the surfaces of a real part differ from them. A film of oil takes mild steel on mild steel from 0.57 to between 0.09 and 0.19 in sliding, and oxide, a coating, wear or a fingerprint changes the coefficient as well. Use the table for the order of magnitude, design with a range and check both ends of it, and where the result depends on friction, measure the coefficient for the surfaces in question.
Solving problems with friction
For any problem involving rough surfaces where frictional forces need to be taken into account one needs to determine if the bodies are at rest or sliding. In general there are several types of states in contact points on a body. If we have a friction coefficient \(\mu_s\) and have the situation
\(F_f<\mu_s N\) then static equilibrium is possible since the surfaces can generate the needed friction force that is required for static friction.
\(F_f=\mu_s N\) then static equilibrium is still possible. This is a limit, often we write \(F_{f,\max}=\mu_s N\), to show that this is the maximum friction force, this limit is also known as fully established friction.
\(F_f>\mu_s N\) then static equilibrium is impossible. The surfaces cannot produce the frictional force that is required to keep static equilibrium. The body begins to slide, accelerates and we get dynamic equilibrium \(F_f=\mu_k N\), at this point the problem is transitioned into a dynamic problem and we have \(\Sigma \bm F=m \bm a\).
The basic rule for analysis with friction is that the frictional force-vector \(\bm F_f\) is set to be in the direction that opposes potential movement.
Problems with friction come in two types:
All external forces except friction forces are known and we need to determine if the friction force can balance the external forces. This means that we need to check if \(F_f \leq \mu_s N\) in the contact surface, otherwise we have sliding and the problem is a dynamic problem with \(F_f=\mu_k N\).
Some external forces are unknown and we want to know how much more we can load the body until sliding occurs. This means that we add \(F_i=\mu_s N_i\) to all rough contact points and include these friction equations in the equilibrium equations. Solving for the unknown forces to answer the problem statement.
We show some typical examples involving friction.
Example 1: A box on a slope
Figure 5.6.3: A box of mass \(m\) on a plank tilted the angle \(\theta\). The coefficient of static friction between them is \(\mu_s\).
Determine the maximum angle \(\theta\) for which the box in Figure 5.6.3 will not slide down the slope. The coefficient of static friction between the surfaces is \(\mu_s\).
Solution
Figure 5.6.4: Free body diagram of the box, with \(t\) along and \(n\) normal to the slope.
Since we’re using a CAS, we can establish a normal coordinate system using the tangential direction \(t\) and the normal direction \(n\) to the slope. Once these directions are established we can easily express the force vectors in these directions and pose the equations of equilibrium.
We have equilibrium equations and an additional friction equation \(F_{\max }=\mu_s N\). Then we just solve for as many variables as we have equations.
Tangential direction: The direction vector in is given by
\[ {\def\arraystretch{1.2}\left[\begin{matrix}F_{f} \cos{\left(\theta \right)} - N \sin{\left(\theta \right)}\\F_{f} \sin{\left(\theta \right)} + N \cos{\left(\theta \right)} - g m\\0\end{matrix}\right] = \left[\begin{matrix}0\\0\\0\end{matrix}\right]} \]
Additionally we have the friction equation
ekv2 = sp.Eq(F_f, mu*N)
\[ {\def\arraystretch{1.0}F_{f} = N \mu} \]
All together we have three equations and three unknowns. We can solve for the unknowns using SymPy.
sol = sp.solve([ekv1, ekv2], [F_f, N, mu], dict=True)[0]sol | la
\[ \begin{aligned}
F_{f} &= g m \sin{\left(\theta \right)} \\[9.0pt]
N &= g m \cos{\left(\theta \right)} \\[9.0pt]
\mu &= \tan{\left(\theta \right)}
\end{aligned} \]
Here we have that \(\theta = \tan^{-1}(\mu_s)\), also known as the angle of friction.
Note
A note for SymPy solve function. There are a number of ways solve returns solutions. For unique solutions we expect a list of one element containing a dict. Sometimes we just get the dict. One needs to check the output and use correct syntax to extract the solution.
In the above case we get a list with one element, which is a dict. First we need to index into the list and grab index 0 then we have the dict.
Note also that we explicitly use dict=True to get a dict as output. This should be the default behaviour of SymPy. But not always. In this case for unknown reasons, if omitted, we would get a list of solutions instead and not a dict.
Example 1 read backwards is a measurement. The box slips at \(\theta = \tan^{-1}\mu_s\), so raising the plank of Figure 5.6.3 slowly from horizontal until the box starts to slide, and reading the angle \(\theta_s\), gives
\[
\mu_s = \tan\theta_s
\tag{5.6.2}\]
The mass drops out, and we only need to measure the angle to get the friction coefficient. Lowering the plank again until the box, given a tap, slides down at constant speed gives the kinetic coefficient the same way, \(\mu_k = \tan\theta_k\), since at constant speed the box is again in equilibrium, now with \(F_f = \mu_k N\).
The pull test measures the force itself (Figure 5.6.5 (a)). A string pulls the block along a fixed surface through a force gauge, a spring balance or a load cell, at a slow and steady speed \(v\). With the string horizontal, \(N = mg\), and the gauge reads the friction force. The reading rises while the block sticks, peaks at \(\mu_s N\) when it breaks loose, and drops to the sliding friction \(\mu_k N\), around which it wavers as the block sticks and slips (Figure 5.6.5 (b)). Repeating the test with added weights gives pairs of \(N\) and \(F_f\). The slope of the straight line through them is the coefficient, and points that bend away from a straight line show that Coulomb’s law does not hold for that pair.
(a) The test. The gauge pulls the block of mass \(m\) at the speed \(v\) and reads the friction \(F_f\).
(b) The gauge reading against time, schematic.
Figure 5.6.5: The pull test. The peak of the reading is the static friction \(\mu_s N\), the level after it the kinetic friction \(\mu_k N\).
Where Coulomb’s law ends
Coulomb’s law gives dry friction one number per pair of surfaces, and that number depends on more than the law shows. For a polymer on steel, the friction force does not even grow in proportion to the normal force. Mokha, Constantinou and Reinhorn measured PTFE sliding on polished stainless steel, the bearing material of bridges and of buildings on seismic isolators, and found that the coefficient of friction falls as the contact pressure rises, levelling off only above about 34 MPa [6]. Doubling the load on such a bearing less than doubles its friction, and the points of a pull test on PTFE bend toward a flatter line as the weights grow. The same tests showed the coefficient rising with the sliding speed up to about 0.1 to 0.2 m/s.
Temperature, lubrication and wear change the coefficient as well. They belong to tribology, the study of friction, wear and lubrication, in which Coulomb’s law is the simplest model; the further reading follows them there. For statics and for most of the machine elements in this book the law is enough, provided the coefficient comes from a test under conditions close to those of the design.
Example 2: Sliding boxes
Figure 5.6.6: The mass \(m_1\) on a slope of angle \(\theta\), tied over a pulley to the hanging mass \(m_2\).
Two masses are rigged as in Figure 5.6.6. In what interval does \(m_2\) need to be in to prevent sliding of \(m_1\). Let \(\mu_s=0.3\) and \(\theta=20^{\circ}\).
Solution - Reformulation of the problem
We need to ensure that we understand the problem. This is imperitive to solve any engineering problem correctly. If in doubt, ask the problem owner for a clarification.
Let’s break down the above problem statement:
We \(m_2\) need to find the interval of \(m_2\) such that \(m_1\) does not slide. This means that we need to find the maximum and minimum values of \(m_2\) such that \(m_1\) is in static equilibrium. In other words:
Find
\[
m_{2, \min} < m_2 < m_{2, \max}
\]
such that \(m_1\) is in static equilibrium.
This gives rise to two cases: depending on if the box is just about to start gliding up or down. These cases result in two FBDs where the direction of the friction force is opposing the direction of movement.
Case 1: \(m_1\) is just about to start sliding down, i.e., finding \(m_{2\min}\)
Figure 5.6.7: Free body diagram of \(m_1\) about to slide down. The friction \(F_f\) acts up the slope, along with the rope force \(T\).
An assumption we make is that the force \(T\) is transmitted through the pulley without any losses. This means that the tension in the rope is constant. We can also assume that the mass of the rope is negligible. This means that the tension in the rope is constant and equal to \(m_2 g\).
Kinematics
We need to establish the directions \(t\) and \(n\). Starting with the tangential direction \(t\) we have
\[
\bm e_t = [\cos \theta, \sin \theta, 0]^\mathsf T
\] The normal direction is given by \[
\bm e_n = [\cos(\theta + 90^\circ), \sin(\theta + 90^\circ), 0]^\mathsf T
\]
Figure 5.6.8: The least mass ratio \(m_2/m_1\) that keeps \(m_1\) from sliding down, against the slope angle, for \(\mu_s = 0.3\). Below zero no rope force is needed: friction alone holds the box.
m_2_min_fun = sol[m_2].trigsimp().subs({mu:0.3})m_2_min_fun | la
Feasibility analysis: At close to \(90^{\circ}\) the frictional force is so small that we get \(m_1=m_2\). At zero degrees the model does not work since we would get negative masses. Case 1 does not exists below slopes of ca \(17^{\circ}\), i.e., the box cannot slide downwards if the slope is less than \(17^{\circ}\).
Case 2: \(m_1\) is just about to start sliding up, i.e., finding \(m_{2\max}\)
Figure 5.6.9: Free body diagram of \(m_1\) about to slide up. The friction \(F_f\) now acts down the slope.
In this case we have the following forces, with the friction force being reversed
Figure 5.6.10: The largest mass ratio \(m_2/m_1\) before \(m_1\) slides up, against the slope angle, for \(\mu_s = 0.3\).
Conclusion
The interval is given by
\[ 0.06012 m_{1} < m_2 < 0.6239 m_{1} \]
Example 3 - Ladder
Figure 5.6.11: A homogeneous ladder of length \(L\) standing on the ground at \(A\) and leaning on a frictionless roller at \(O\), a height \(L/2\) above the ground. The angle \(\theta\) is measured from the vertical.
Determine the least possible \(\mu_s(\theta)\) at \(A\) such that the homogeneous ladder of mass \(m\) in Figure 5.6.11 does not slide. There is no friction at \(O\). Disregard the radius of the roller at O. Draw a graph of \(\mu_s(\theta)\) for valid angles.
Solution
Figure 5.6.12: Free body diagram of the ladder. The roller pushes with \(N_O\) along the normal \(n\) to the ladder, the ground with \(N_A\) and the friction \(F_f\), and the weight \(F_G\) acts at the midpoint.
We start by defining the kinematics, the position of \(A\) needs to be defined as a function of that angle \(\theta\). Using trigonometry we recognize the triangle and have
Figure 5.6.14: The distance from \(A\) to the roller, \(\lvert\bm r_{OA}\rvert\), in units of the ladder length \(L\). Above 1 the roller is beyond the top end of the ladder.
As Figure 5.6.14 shows, we cannot go beyond \(60^\circ\), which the CAD sketch in Figure 5.6.15 confirms.
Figure 5.6.15: A CAD sketch of the ladder, length 100 on a roller at height 50. Turning the ladder changes \(\theta\), and at \(\theta = 60^\circ\) the roller reaches the top end, since \((L/2)/\cos\theta = L\).
Example 4 - Flexible belts (Capstan equation)
Ropes and belts wrapped around cylinders are a common application of static friction in mechanics. The sheet winch in Figure 5.6.16 is one: a few turns of rope around the drum let a small hand force hold the large load in the sheet. We shall look closer at the problem.
Figure 5.6.16: A sheet winch on a sailing boat. A few turns of rope around the drum let a small hand force hold the large load in the sheet.
Free body diagram
Infinitesimal belt element
Figure 5.6.17: A rope wrapped the angle \(\beta\) around a drum. Left: the drum with the rope tensions \(T_1\) and \(T_2\). Right: a rope element \(d\theta\) with the normal force \(dN\) and the friction force \(\mu\,dN\).
A rope is wrapped some angle \(\beta \in[0, \infty]\) radians around a cylinder. We cut a small piece \(d \theta\) at an arbitrary angle \(\theta \in[0, \infty]\) and create a FBD as in Figure 5.6.17. We are interested in the maximum force and examine the limit just before sliding. The force equilibrium is divided into a tangential \((t)\) and radial \((n)\) direction
Since both dT and \(d \theta\) are very small, their product is even smaller and can be neglected, i.e., \(dTd \theta \ll dT, d \theta\)
Thus we end up with
\[
\left\{\begin{array}{l}
\mu dN=dT \\
dN=T d \theta
\end{array}\right.
\]
Eliminating \(dN\) and moving things around we can formulate a simple ordinary differential equation
\[
\frac{dT_2}{d \theta}=\mu T_2
\]
with the boundary condition \(T_2(0)=T_1\)
For which we can easily get the solution using SymPy symbolic differential equation solver dsolve:
import sympy as spfrom sympy import difftheta, mu, T1 = sp.symbols('theta, mu, T1', real=True)# Here we define the function T2T2 = sp.Function('T2')T2(theta)
\(\displaystyle T_{2}{\left(\theta \right)}\)
# Define the Differential EquationDE = sp.Eq(diff(T2(theta), theta), mu * T2(theta))DE
Ok, so with, e.g., \(\mu_s=0.25\) (hamp against steel according to some table) and wrapped five times we get
from sympy import exp, pimu =0.25T2 = T1 * exp(mu*5*2*pi)T2.evalf(6)
\(\displaystyle 2575.97 T_{1}\)
We see that the force \(T_2\) needs to be over 2500 times larger than \(T_1\) to be in balance. The Capstan equation (5.6.3) is extremely useful in many applications, e.g., belt breaking, winches, cranes, gearboxes, etc.
For a cool application see this video by Aaed Musa:
Example 5 – Clutch
A clutch transfers torque between two rotating shafts through friction. Two flat plates are pressed together by an axial force \(P\), and the friction between the contact surfaces transmits a moment \(M\). We want to relate the transmittable moment to the applied force, the friction coefficient, and the radius of the clutch face. Figure 5.6.18 shows the two plates on their shafts and, seen along the axis, the ring element of the contact face that the integration below is built on.
Figure 5.6.18: The clutch. Left: the two plates pressed together by \(P\) and turned by \(M\). Right: the contact face with an element of area \(dA = r\,d\theta\,dr\) at radius \(r\), on which the friction force \(\mu p\,dA\) acts tangentially.
We slice the face into onion rings. A ring at radius \(r\) and width \(dr\) has area \(dA = 2\pi r\,dr\). With a uniform contact pressure \(p\), it carries the normal force \(p\,dA\) and the friction force \(\mu p\,dA\), which has the lever arm \(r\) about the axis. Over a face from \(R_i\) to \(R_o\),
\[
P = \int_{R_i}^{R_o} p\,2\pi r\,dr, \qquad M = \int_{R_i}^{R_o} r\,\mu p\,2\pi r\,dr
\]
and we eliminate \(p\) to get \(M\) in terms of \(P\).
r, R_i, R_o, p, P, M = sp.symbols('r R_i R_o p P M', positive=True)mu = sp.Symbol('mu', positive=True)# Contributions of one onion ring at radius rdA =2*sp.pi*rdP = p*dAdF = mu*dPdM = r*dF{'dA': dA, 'dP': dP, 'dF': dF, 'dM': dM} | la
\[ \begin{aligned}
dA &= 2 \pi r \\[9.0pt]
dP &= 2 \pi p r \\[9.0pt]
dF &= 2 \pi \mu p r \\[9.0pt]
dM &= 2 \pi \mu p r^{2}
\end{aligned} \]
Code
# Integrate from R_i to R_oP_total = sp.integrate(dP, (r, R_i, R_o))M_total = sp.integrate(dM, (r, R_i, R_o)){sp.Symbol('P'): P_total, sp.Symbol('M'): M_total} | la
\[ \begin{aligned}
P &= - \pi R_{i}^{2} p + \pi R_{o}^{2} p \\[9.0pt]
M &= - \dfrac{2 \pi R_{i}^{3} \mu p}{3} + \dfrac{2 \pi R_{o}^{3} \mu p}{3}
\end{aligned} \]
Eliminating \(p\):
Code
sol = sp.solve([sp.Eq(P_total, P), sp.Eq(M_total, M)], [M, p])sol | la
\[ \begin{aligned}
M &= \dfrac{2 P R_{i}^{2} \mu + 2 P R_{i} R_{o} \mu + 2 P R_{o}^{2} \mu}{3 R_{i} + 3 R_{o}} \\[9.0pt]
p &= - \dfrac{P}{\pi R_{i}^{2} - \pi R_{o}^{2}}
\end{aligned} \]
For a full disc, \(R_i = 0\) and \(R_o = R\), this reduces to \(M = \tfrac{2}{3}\mu P R\) and \(p = P/(\pi R^2)\). The moment scales linearly with \(\mu\), \(P\) and the radius. Real clutches use multiple friction surfaces in series to multiply \(M\) without enlarging \(R\).
Example 6 – Screws
A screw converts rotation into linear motion (or, depending on use, transfers an axial load through a rotational input). A trapezoidal screw maximizes efficient transfer of torque to linear force, while a standard machine screw is shaped to maximize axial load capacity. Figure 5.6.19 shows the two profiles side by side and Figure 5.6.20 the section of a trapezoidal screw in its nut. Regardless of profile the analysis is the same: unwrap one revolution of the thread into a straight inclined plane, and the problem reduces to a block on a slope under friction.
Figure 5.6.19: A trapezoidal thread and a metric thread. The trapezoidal profile carries a coarse pitch and transfers torque into axial motion efficiently, the metric profile a fine pitch and a larger load capacity.
Figure 5.6.20: Section through a trapezoidal screw and its nut. The pitch \(p\) is the distance between neighbouring crests and \(r\) is the mean radius, halfway between the crest and the root of the thread.
Lead screws with a trapezoidal thread drive the axes of 3D printers and other small machines. In Figure 5.6.21 the motor turns the screw, and the brass nut, held in a printed block that runs on a linear guide, carries the load along the axis.
(a) The nut in its printed block on the guide rail.
(b) The stepper motor that turns the screw.
Figure 5.6.21: A lead screw driven by a stepper motor.
The lead \(l\) is the height the screw advances per revolution, and for the single-start threads we consider it equals the pitch. In Figure 5.6.22 the points \(A\) and \(B\) sit on the same thread one turn apart, so \(B\) is the lead \(l\) above \(A\). Unwrapping that turn along the mean radius, Figure 5.6.23, gives a triangle whose hypotenuse is the thread, whose horizontal leg is \(2\pi r\) (the circumference at the mean radius), and whose vertical leg is \(l\). The thread angle \(\theta\) above the horizontal is \(\tan\theta = l/(2\pi r)\).
Figure 5.6.22: One turn of thread. \(B\) lies on the thread one revolution after \(A\), a lead \(l\) higher up the screw of mean radius \(r\).
Figure 5.6.23: The turn from \(A\) to \(B\) unwrapped into a wedge of base \(2\pi r\) and height \(l\), inclined at the thread angle \(\theta\).
Trapezoidal screws have a larger pitch than standard machine screws, which makes them more efficient at converting rotation into linear motion at the cost of holding capacity. Ball screws reduce friction further with rolling contact at the cost of self-locking: an axial load can easily back-drive a ball screw, so a motor torque is required to hold position. Even trapezoidal screws can be back-driven given enough axial force. Machine screws on the other hand are self locking and cannot be back-driven. Pitch values for trapezoidal screws are tabulated by suppliers, e.g. mekanex.
Mechanical analysis
We isolate a small segment of the thread and treat it as a block on the unwrapped incline. The thread angle \(\theta\) is the helix angle: \(\tan\theta = \text{pitch}/(2\pi r)\). The friction angle \(\alpha\) is defined by \(\tan\alpha = \mu\), so the contact reaction \(\bm R = \bm N + \bm F_f\) makes angle \(\alpha\) with the surface normal whenever the block is on the verge of sliding.
Three situations matter:
Raising a load: we apply torque to push the screw upward against the load.
Lowering when \(\theta < \alpha\): the screw is self-locking, the load alone cannot drive it. To lower, we must apply torque in the opposite direction.
Self-unwinding when \(\theta > \alpha\): the thread is steep enough that the load alone overcomes friction. To hold the load we must apply torque in the lifting direction.
In all cases the analysis is the same FBD with three forces \(\bm W\), \(\bm F_f\), \(\bm N\) and an applied tangential force \(M/r\). Only the direction of friction (and of \(M/r\)) changes between cases.
Raising the load
Figure 5.6.24: Raising the load. Friction acts down the incline and the tangential force \(M/r\) pushes the block up it. The reaction \(\bm R\) is \(\bm N\) and \(\bm F_f\) combined, tilted the friction angle \(\alpha\) from the normal \(n\).
In Figure 5.6.24 the block slides up the incline, friction \(F_f = \mu N\) acts down the incline opposing the motion, and we push horizontally with \(M/r\) in the raising direction. With \(x\) horizontal in the raising direction and \(y\) up, the two components of force equilibrium and Coulomb’s law at the verge of sliding give three equations for the three unknowns \(M\), \(N\) and \(F_f\),
The torque needed to raise the load increases with both the thread angle (steeper thread, more axial advance per turn) and the friction angle (more friction, more loss). The denominator \(\cos(\theta+\alpha)\) vanishes when \(\theta + \alpha = \pi/2\), the geometric limit where any further increase in either angle would require infinite torque to lift.
Lowering the load when \(\theta < \alpha\)
Figure 5.6.25: Lowering the load when \(\theta < \alpha\). Friction acts up the incline, and the reaction \(\bm R\) leans past the vertical, so a tangential force in the lowering direction is needed to move the block.
In Figure 5.6.25 the thread angle is smaller than the friction angle, so friction alone is enough to hold the load. The screw is self-locking: it will not unwind on its own. To make the load go down we must actively push the screw in the lowering direction. Friction now acts up the incline (opposing the lowering motion), and the applied tangential force \(M/r\) points the other way. Both terms change sign in the horizontal equation, and friction changes sign in the vertical one,
Since \(\alpha > \theta\) in the self-locking regime this is positive: a real torque is required in the lowering direction to break the friction and bring the load down.
Self-unwinding when \(\theta > \alpha\)
Figure 5.6.26: Holding the load when \(\theta > \alpha\). Friction acts up the incline but cannot hold the block alone, so the tangential force \(M/r\) must act in the raising direction.
In Figure 5.6.26 the thread angle exceeds the friction angle, so the geometry tips the other way: the axial load alone overcomes friction and the screw unwinds on its own. To hold the load steady we must apply torque in the lifting direction. The block is on the verge of sliding down, so friction acts up the incline as in the lowering case, but now \(M/r\) has to act in the raising direction to prevent motion. Compared with the lowering case only the sign of \(M/r\) changes,
\[ {\def\arraystretch{1.0}- W r \tan{\left(\alpha - \theta \right)}} \]
SymPy reports this as \(-Wr\tan(\alpha - \theta)\), which is the same as \(Wr\tan(\theta - \alpha)\) but written with \(\alpha - \theta\) inside the \(\tan\) so the symbolic expression follows the same pattern. Either way
is positive precisely when \(\theta > \alpha\), the self-unwinding regime. The three boxed formulas, 5.6.4, 5.6.5 and 5.6.6, summarize the entire analysis: only the sign and the order of \(\theta\) and \(\alpha\) in \(\tan(\theta \pm \alpha)\) change between the three situations.
Standard machine screws have small thread angles (a flat helix and a small pitch) well below typical friction angles, so they are self-locking and hold a load by themselves. Trapezoidal screws have larger thread angles and tend toward self-unwinding under heavy axial loads. Ball screws have very low friction angles (rolling contact) and back-drive easily, which is why they need a motor torque to hold position.
Example 7 – A servo-controlled brake
Figure 5.6.27: A brake for the shaft at \(E\). The servo turns the arm \(AB\), the spring \(k\) pulls \(C\) towards \(B\), and the grey arm, which turns about the screw at \(D\), presses on the shaft. Lengths in mm.
In Figure 5.6.27 a servo turns the arm \(AB\) of length \(r = 30\) mm. A spring of stiffness \(k = 100\) N/mm and free length \(L_0 = 40\) mm joins \(B\) to the end \(C\) of a printed brake arm. The brake arm turns about the screw at \(D\) and has a hole around the shaft at \(E\), of diameter \(d = 8\) mm. When the spring is stretched it pulls the brake arm towards the servo, the edge of the hole presses on the shaft, and the friction between them, with \(\mu = 0.35\), brakes the shaft. Determine the braking moment on the shaft at \(E\) as a function of the angle \(\theta\) of the arm \(AB\).
Solution
We put the origin at \(A\), with \(x\) to the right and \(y\) towards \(E\), and measure \(\theta\) from the \(x\) axis, positive anticlockwise; in the photo the arm points a few degrees below the axis. From the dimensions in Figure 5.6.27, in mm,
An extension spring can only pull. When the servo turns \(B\) so far up that \(L < L_0\), the spring hangs slack, \(F_s = 0\) and the brake is off.
Figure 5.6.28: The brake arm freed from the pin at \(D\) and from the shaft, drawn for \(\theta = -10^\circ\). The dashed circle is the shaft, which touches the edge of the hole at \(K\).
Figure 5.6.28 frees the brake arm. The spring pulls \(C\) towards \(B\), the arm turns a little about \(D\), and the upper edge of the hole comes to rest on the top of the shaft at \(\bm r_K = \bm r_E + [0,\ d/2]^\mathsf{T}\). There the shaft pushes on the arm with the normal force \(N\), along \(y\) through the centre of the shaft, and with the friction force along \(x\). The shaft slips under the arm, so friction has its full value \(\mu N\), and its direction follows the shaft: we write it \(s\mu N\), with \(s = +1\) when the shaft turns clockwise in the photo and \(s = -1\) when it turns anticlockwise. The pin force at \(D\) is unknown, and the moment equation about \(D\) leaves it out:
With \(\bm r_C - \bm r_D = [45,\ 0]^\mathsf{T}\) and \(\bm r_K - \bm r_D = [20,\ d/2]^\mathsf{T}\) the \(z\) components give
\[
\begin{aligned}
45\,F_{s,y} + 20\,N - \frac{d}{2}\,s\mu N &= 0\\
N &= \frac{-45\,F_{s,y}}{20 - s\mu d/2}
\end{aligned}
\]
where \(F_{s,y}\) is negative, since the spring pulls \(C\) towards the servo. The friction force \(\mu N\) acts on the shaft at the radius \(d/2\), so the braking moment is
\[
M_E = \mu N\,\frac{d}{2}
\tag{5.6.7}\]
Code
import numpy as npimport sympy as spimport matplotlib.pyplot as plt%config InlineBackend.figure_formats = ['svg']from mechanicskit import la, ltxtheta, s = sp.symbols('theta s', real=True) # arm angle; s = +1 or -1, the shaft's senseN = sp.symbols('N', positive=True)k, L_0, r, d =100, 40, 30, 8# N/mm, mm, mm, mmmu = sp.Rational(35, 100)rr_D = sp.Matrix([-20, 40, 0])rr_E = sp.Matrix([0, 40, 0])rr_C = sp.Matrix([25, 40, 0])rr_K = rr_E + sp.Matrix([0, sp.Rational(d, 2), 0]) # top of the shaft, where the hole touchesrr_B = r*sp.Matrix([sp.cos(theta), sp.sin(theta), 0])L = (rr_C - rr_B).norm()F_s = k*(L - L_0) # holds while the spring is stretched, L > L_0FF_s = F_s*(rr_B - rr_C)/L # the spring pulls C towards BFF_K = sp.Matrix([s*mu*N, N, 0]) # the shaft on the arm: friction and normal forcemoment_D = (rr_C - rr_D).cross(FF_s)[2] + (rr_K - rr_D).cross(FF_K)[2]N_sol = sp.solve(sp.Eq(moment_D, 0), N)[0]M_E = mu*N_sol*sp.Rational(d, 2) # braking moment on the shaft [N mm]th_photo = sp.rad(-10)at_photo = {sense: {q: float(expr.subs({theta: th_photo, s: sense}))for q, expr in (('L', L), ('F_s', F_s), ('N', N_sol), ('M_E', M_E))}for sense in (1, -1)}
In about the position of the photo, \(\theta = -10^\circ\), the spring is stretched by 5.4 mm, and the two senses of rotation give
Turned clockwise, the shaft drags the arm in the direction the spring already turns it, and the arm presses harder: the brake is self-energizing in that direction. The effect is small here, about 7 % either way, since the friction acts only \(d/2 = 4\) mm from the line \(DE\). A brake shoe whose pivot sits far off the friction line, as in a drum brake, gains much more, and if \(s\mu d/2\) reached 20 the denominator would vanish and the brake would lock.
Figure 5.6.29 shows \(M_E\) over the angles the servo can reach. Below it, as a check of whether the servo can hold the arm, is the moment the spring puts on the servo, \(M_A = [\bm r_B \times (-\bm F_s)]_z\), against the 25 kg cm \(= 2.45\) N m of the servo in the photo.
Code
M_A = rr_B.cross(-FF_s)[2] # the spring's moment on the servo arm [N mm]slack = sp.lambdify(theta, L - L_0)f_E = sp.lambdify((theta, s), M_E)f_A = sp.lambdify(theta, M_A)deg = np.linspace(-135, 30, 661)th = np.radians(deg)on = slack(th) >0# the spring is stretched; slack means no forceM_servo =25*9.81*0.01# 25 kg cm [N m]fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(6.4, 6.4), sharex=True)for sense, ls, lab in ((1, '-', r'$s=+1$, clockwise'), (-1, '--', r'$s=-1$, anticlockwise')): ax1.plot(deg, np.where(on, f_E(th, sense), 0)/1e3, ls, color='#236B8E', label=lab)ax1.set_ylabel(r'$M_E$[N m]')ax1.set_ylim(0, 12)ax1.set_yticks(np.arange(0, 13, 3))ax1.legend(frameon=False)ax2.plot(deg, np.where(on, f_A(th), 0)/1e3, color='#c00000')ax2.axhline(M_servo, color='0.4', lw=0.9, ls=':')ax2.text(-132, M_servo +1.5, 'servo, 2.45 N m', color='0.3', fontsize=9)ax2.set_ylabel(r'$M_A$[N m]')ax2.set_ylim(-20, 60)ax2.set_yticks(np.arange(-20, 61, 20))ax2.set_xlabel(r'$\theta$[$^\circ$]')ax2.set_xlim(-135, 30)ax2.set_xticks(np.arange(-135, 31, 15))for ax in (ax1, ax2): ax.grid(alpha=0.3)plt.tight_layout()plt.show()
Figure 5.6.29: Top: the braking moment 5.6.7 for the two senses of rotation. Bottom: the moment the spring puts on the servo arm, and the dotted line the servo can hold. The spring is slack to the right of \(\theta \approx 0.6^\circ\).
Code
th_free =float(sp.nsolve(L - L_0, theta, 0.0)) # spring at its free lengthth_servo =float(sp.nsolve(M_A - M_servo*1e3, theta, -0.02)) # servo at its limitth_toggle =float(sp.atan2(-40, -25)) # A on the line BCM_E_servo =float(M_E.subs({theta: th_servo, s: 1}))/1e3
Both curves pass the check that they are zero where the spring reaches its free length, and the braking moment grows as the servo turns \(B\) down and stretches the spring, up to a largest value near \(\theta = -103^\circ\), where \(B\) starts to rise again and the spring leans over towards the direction of \(DC\), so its moment about \(D\) falls although it still stretches. The servo check is what decides the design. With \(k = 100\) N/mm the servo holds the arm only down to
so the braking moments of several N m in the upper plot are out of reach: the spring would turn the servo back. Since both \(M_E\) and \(M_A\) are proportional to \(k\), a softer spring does not help, it only moves the limit to a larger angle, and the ratio \(M_E/M_A\) that the geometry sets is what the servo’s 2.45 N m buys. The lower plot also shows a way out. At \(\theta = \operatorname{atan2}(-40, -25) \approx -122^\circ\), \(A\) lies on the line \(BC\), the spring force passes through the servo axis and \(M_A = 0\), while the spring is stretched almost to its longest. A servo parked in that position holds a large braking moment for free, as a toggle clamp does; past it the moment changes sign and the spring pulls the arm on against a stop.
Friction in machine elements
The examples of this chapter return in the Machine Elements part, where friction carries the load. The thread of Example 6 becomes the bolted joint, whose preload is set by the friction in the thread and under the head. The capstan of Example 4 becomes the belt drive, where the tension ratio of 5.6.3 limits the torque a belt transmits before it slips. The clutch of Example 5 and the band of Example 4 return in brakes and clutches, and the clamped hub and the shrunk-on gear of shaft-hub joints transmit their torque by friction alone.
Further reading
Tribology is the study of friction, wear and lubrication between surfaces in relative motion, and Hutchings and Shipway [7] introduce it for engineers. It explains friction from the contact mechanics of the real area of contact, the small fraction of the apparent area where the asperities of the two surfaces touch. Bowden and Tabor [8] showed that this area grows with the normal force because the asperities flatten under it, and that the friction force follows the real area, so Amontons’ proportionality is a consequence of how surfaces deform and not a law of nature. Persson [9] treats sliding friction from its physical principles, from the atomic scale to rubber and to earthquakes, where a fault slides under a friction law that depends on the slip rate and on the history of the contact [10]. Popov [11] joins contact mechanics and friction with the numerical methods that solve them, and Dowson [12] follows the history of the field from the sledges of ancient Egypt through Leonardo, Amontons and Coulomb to the twentieth century.
Temperature changes the coefficient. The friction of PTFE on steel is unaffected at moderate temperatures but rises below freezing, and the design rules for PTFE bearings multiply the coefficient by a factor rising from 1.0 at 20 °C to 2.0 at −50 °C [13]. Heat lowers it: a car brake whose pads run hotter than their working temperature grips less, which is known as fade and is treated in the brakes chapter.
A film of oil between the surfaces makes friction depend on the speed and the viscosity, as Stribeck measured in journal bearings [14]. The pressure in the film follows from Reynolds’ equation for the thin flow between the surfaces [15]. Bertil Nilsson and Peter Hansbo developed adaptive finite element methods for that flow, including the cavitation where the film breaks up [16], and coupled Reynolds’ thin-film model to the full Stokes equations where the gap is not thin [17,18].
Wear is modelled in the same spirit as friction. According to Archard [19], the volume \(V\) worn away grows in proportion to the normal force \(N\) and the sliding distance \(s\), and in inverse proportion to the hardness \(H\) of the softer surface, \(V = K N s/H\), with a wear coefficient \(K\) that, like \(\mu\), is measured for a pair of surfaces. Combined with the finite element method, the law predicts how a contact changes shape over its life [20]: the contact pressure of a finite element solution gives the local wear over a small step of sliding, the worn surface is moved by that depth, and the contact is solved again. Strömberg developed an augmented Lagrangian method for frictional contact in fretting, the small oscillating slip that wears joints apart [21]; the method enforces the contact conditions exactly while keeping the equations well conditioned. With Rashid he simulated a disc brake over repeated stops, updating the wear of the pad after each one, and predicted the hot bands that form on the disc [22].
References
[1]
Baumeister T, Avallone EA, Baumeister III T, editors. Marks’ standard handbook for mechanical engineers. 8th ed. New York: McGraw-Hill; 1978.
DIN 7190-1:2017 Interference fits – Part 1: Calculation and design rules for cylindrical self-locking pressure fits. Berlin, Germany: Deutsches Institut für Normung; 2017.
Stribeck R. Die wesentlichen Eigenschaften der Gleit- und Rollenlager. Zeitschrift Des Vereines Deutscher Ingenieure 1902;46.
[15]
Reynolds O. On the theory of lubrication and its application to Mr.BeauchampTower’s experiments. Philosophical Transactions of the Royal Society of London 1886;177:157–234.
[16]
Nilsson B, Hansbo P. Adaptive finite element methods for hydrodynamic lubrication with cavitation. International Journal for Numerical Methods in Engineering 2007;72:1584–604.
[17]
Nilsson B, Hansbo P. Weak coupling of a Reynolds model and a Stokes model for hydrodynamic lubrication. International Journal for Numerical Methods in Fluids 2011;66:730–41.
[18]
Nilsson B, Hansbo P. A Stokes model with cavitation for the numerical simulation of hydrodynamic lubrication. International Journal for Numerical Methods in Fluids 2011;67:2015–25.
[19]
Archard JF. Contact and rubbing of flat surfaces. Journal of Applied Physics 1953;24:981–8.
[20]
Põdra P, Andersson S. Simulating sliding wear with finite element method. Tribology International 1999;32:71–81.
[21]
Strömberg N. An augmented Lagrangian method for fretting problems. European Journal of Mechanics A/Solids 1997;16:573–93.