Active real-time learning on a test rig for pressure-spike mitigation in digital hydraulic systems

Suppressing rapid pressure spikes is important for safe and reliable operation of hydraulic systems. The problemarises especially in digital hydraulic valve systems. This work presents an on-rig training and testing method-ology that applies a context-aware, one-step reinforcement learner to tune relative valve timing and limitpressure spikes. This method bypasses incomplete physics-based models and uses only limited measurements, with no large, labeled dataset. It also avoids extra passive hydraulic hardware. The test environment is a high-capacity digital flow control unit prototype delivering 480 LPM at Δp ≈ 5 bar, with fast transitions around 25 msand flow steps of about 10 LPM. The semi-binary on/off manifold generates thousands of states for the controllerto evaluate. Real test benches are inherently stochastic: sensor noise, supply fluctuations, and finite bandwidth in thepressure-control devices that challenge many optimization methods. The controller must adapt under realisticuncertainty rather than rely on a deterministic setup. Across representative valve transitions, timing-only ad-justments significantly reduced the surge impulse and, in some cases, eliminated pressure peaks entirely. Thisshows that digital hydraulics combined with active intelligent control can deliver fast, stable, and energy-efficient flow regulation. (PDF) Active real-time learning on a test rig for pressure-spike mitigation in digital hydraulic systems. Available from: https: //www. researchgate. net/publication/401023742Activeᵣeal-timeₗearningₒnₐₜestᵣigforₚressure-spikeₘitigationᵢndigitalₕydraulicₛystems#fullTextFileContent accessed Mar 03 2026.