Method and system for identifying time-varying characteristics of heavy-load vehicle suspension

1 . A method for identifying time-varying characteristics of heavy-load vehicle suspension, comprising a non-transitory computer readable medium operable on a computer with memory for the method for identifying time-varying characteristics of heavy-load vehicle suspension, and comprising program instructions for executing the following steps of:

a). collecting, using sensors, a sequence state of mining truck operations;

b). predicting, using a deep learning network, one or more parameter-related factors based on the sequence state;

c). estimating, using a linear dynamics model considering longitudinal-vertical coupling effects, stiffness and damping coefficients as key suspension parameters;

d). predicting, using a nonlinear dynamics model considering longitudinal-vertical coupling effects, a predicted state at a next time step based on the one or more parameter-related factors and the key suspension parameters;

e). updating network parameters of the deep learning network using a state loss Lp between the predicted state and a target state detected by the plurality of sensors serves as the loss error term;

wherein the linear dynamics model considering longitudinal-vertical coupling effects comprises a vehicle body with mass m c and a pitch moment of inertia Ic, and a front axle and a rear axles with unsprung masses m tf and m tr respectively;

wherein suspension forces transmitted to the vehicle body from the front axle and the rear axles are F f and F r , and road excitations for front and rear tires are z qf and z qr ;

wherein degrees of freedom (DoF) of the linear dynamics model include a vertical displacement of a center of gravity (CoG)z c , a pitch angle θ c , and vertical displacements of the unsprung masses of the front axle and the rear axle z tf and z tr ;

wherein F c represents an inertial force acting on the CoG due to an acceleration a and a velocity v along an x-axis, and M c represents a moment of inertia;

wherein the suspension stiffness and damping coefficients are k and c, with subscripts f and r representing front and rear suspensions, and a subscript t representing a tire;

wherein distances from the CoG to the front axle and the rear axles are a c and b c respectively, and a vertical distance between the CoG and a pitch center PC is Δh; and

f). improving vehicle stability under heavy-duty and off-road conditions while prolonging suspension system longevity based on the identifying time-varying characteristics of the heavy-load vehicle suspension.

2 . The method according to claim 1 , wherein the deep learning network comprising a three-layer Long Short-Term Memory (LSTM) network and a fully connected network layer, wherein each layer of the three-laver LSTM has 128 hidden units, and wherein the fully connected network layer is configured to output suspension stiffness and damping correction coefficients.

3 . The method according to claim 1 , wherein a vertical acceleration and a vertical velocity at a position of the CoG are calculated based on a nonlinear mapping relationship of an Inertial Measurement Unit (IMU), the calculating comprising:

(i) decomposing, according to an Euler theorem of rigid body kinematics, a motion of a point in an absolute coordinate system into a translational displacement in a relative coordinate system plus a vector sum of rotation around a base point, wherein an absolute reference frame is fixed on a ground, a relative coordinate system is fixed on the IMU, and a coordinate system definition corresponds to the IMU itself;

(ii) determining a rotation order of coordinate axes and calculating a rotation matrix by first rotating around a Z-axis corresponding to a yaw angle α c , then rotating around a Y-axis corresponding to the pitch angle θ c and finally rotating around an X-axis corresponding to a roll angle γ c ; and

(iii) calculating a speed of the point and taking a derivative of expressions of the rotation matrix with respect to time to calculate an acceleration, including an angular acceleration, of any point rigidly connected to an installation position of the IMU to obtain a motion of the vehicle body.

4 . The method according to claim 1 , wherein estimating the stiffness coefficient and the damping coefficient comprises:

(i) obtaining a vertical displacement of a sprung mass F f according to a transfer relationship from a suspension force to a suspension displacement based on a Laplace transform relationship between force and displacement;

(ii) eliminating a vertical displacement of the unsprung masses and rewriting a system transfer equation into a standard form with limited continuous-time white noise e(t) using a continuous-time system simplified refined instrumental variable method (SRIVC) for system identification based on a measured real-time timing signal;

(iii) low-pass filtering the system transfer equation using a continuous-time state variable filter to obtain pre-filtered time derivatives of an input and an output, defining an instrumental vector to eliminate noise influence, and iteratively calculating a parameter estimate until an error between an identification result and a previous identification result satisfies a predetermined threshold; and

(iv) evaluating an accuracy of the system transfer equation using a normalized root mean square error (NRMSE) and a final prediction error (FPE) as evaluation indicators.

5 . The method according to claim 1 , wherein predicting the predicted state at the next time step comprises:

(i) performing a backward differentiation to obtain a discrete system expression of the front suspension and the rear suspension at a time k, and introducing a forward differentiation of a body DoF to represent a speed of the unsprung masses at a time k−1 to solve for the suspension forces F f and F r transmitted at the time k;

(ii) summarizing the predicted state of the vehicle body according to the forward differentiation to obtain an iterative method of mining truck body parameters in discrete time; and

(iii) updating iterative parameters using the deep learning network to provide refined modeling based on a time series signal.