Learning-augmented input shaping for boom cranes under operational constraints: A real-world implementation

Boom cranes are flexible and underactuated machines in which suspended loads naturally excite lightly damped oscillations, making fast and accurate positioning difficult in industrial cargo-handling operations. Skilled operators often mitigate sway by intuitively shaping their joystick inputs; however, this reliance on individual dexterity is difficult to scale and leads to variability in cycle time and residual oscillations. Input shaping (IS) provides a simple feedforward method to suppress residual sway, yet its industrial adoption remains limited by the perceived maneuver delay associated with the very low sway frequencies typical of long hoisting cables, as well as by the lack of operator-oriented deployment strategies. This paper presents a learning-augmented input shaping framework for boom cranes based on a combined data-driven and physics-consistent multibody modeling approach. Short video recordings of unshaped slewing maneuvers are used to estimate the dominant sway frequency and damping ratio under real operating conditions. These measurements are used to validate a multibody model of the crane implemented in MATLAB/Simscape, which allows systematic exploration of the operational parameter space (hoist length and payload inertia) and provides the frequency envelope that must be attenuated by the shaper. The unshaped maneuver duration T 0 is automatically learned in real time by routing the operator’s joystick command through the cyclic interrupt routine of the PLC, enabling synchronization of the shaped command without external sensing or manual calibration. In practical bulk unloading operations, the learning procedure is typically performed only once for each ship hold compartment and truck hopper configuration, since the corresponding point-to-point payload trajectory remains essentially unchanged during the unloading process. The shaping sequence is generated in the PLC using a time-indexed toggle logic that reproduces the impulse sequence without real-time convolution, reducing computational load. Unlike most studies limited to simulations or laboratory prototypes, the proposed strategy is implemented on an industrial Siemens S7-1511 PLC controlling a Gottwald boom crane operating in the Port of Marín, Spain. Multibody simulations and field observations demonstrate effective reduction of payload oscillations across the operational frequency range while preserving operator workflow and predictable maneuver duration.