**First-step optimization (MATLAB model)**

The first step was to generate the curve with MATLAB and obtain the results for jerk, Hertz pressure and force. Fig.3. shows all the variables used for the creation. The Design variables: X=[xb;xe;xf;ye;y’c;y’d;y’e;y’’c;y’’d;y’’e]; For this part, the value of each variable is not very perceptive physically (eg: the jerk at a point) and the range can be very large eg. +/- 106. The mechanism’s response is multimodal, possibly due to the high number of variables. The constraints on the mechanism were also very difficult to handle; only a small range of combinations of variables can adapt to the constraints. The way to find the best solution is to create a Design of Experiments (DOE) of 30 designs and use a MOGA II as the optimization algorithm with 400 generations to explore the domain. This takes two hours, on average, which is quite fast. The second step is to use an adaptive filter sequential quadratic programming (AFSQP) method on the best design found with MOGA II, which allows one to gain 12% on the cost function.

**Results**

#### Fig. 4 - Hertz pressure generated by the previous and optimized laws of motion

Fig. 4 shows the Hertz pressure. The reduction is 10 MPa, which increases the angular velocity of the cam’s inlet and consequently the productivity of the machine. Fig. 5 shows that the contact force in the second zone increased to 60%, which ensures permanent and reliable contact between the cam and the follower roller. The optimization also leads to continuity of Jerk, which was the third objective of the multicriteria design optimization.

**Second-step optimization (Adam-MATLAB model)**

#### Fig. 5 - Contact force generated by the previous and optimized laws of motion

The classic dynamic model is not precise enough to estimate the maximum force generated on the mechanism. In reality, other phenomena such as vibration or shock may appear. A good way to estimate this maximum value is to run a simulation in Adam. To meet the objective of minimizing the maximum force on the mechanism, we must insert Adam into the modeFRONTIER flow, which requires a little adaptation as depicted in Fig.6. The main idea is to use a .cmd file for Adam with the entire mechanism drawn in it and a special part, which is the cam, created on the basis of a spline defined by a point matrix. This point matrix contains all the points of the cam profile. For each design drawing, we have to create a new .cmd file in which the point matrix is modified with the new curve. The .cmd file has a very rigorous structure, so all the modifications must be well oriented to avoid creating errors in the simulation. Then the simulation is run in Adam to obtain the desired data. The key disadvantage of this approach is that the Adam simulation takes some time to execute so it is useful to limit the time of each simulation in the Adam node. This requires being truly aware of the physical behavior of the mechanism. The algorithm used here is only a MOGA II with 30 DOE and 600 generations.

**Results**

In conclusion, Fig 7 shows that the value of the maximum force has been reduced by 17%, which increases the mechanism’s lifetime by 100 times, on average.

#### Fig. 6 - Creation algorithm for the cam profile

#### Fig. 7 - Flow associated with the Adam view