Machine-learning source localisation in an asymmetric multimode photoacoustic cell

In a multimode photoacoustic cell, the acoustic response depends on the spatial distribution of absorbed power. For heat deposition confined to a short axial region, its position is represented by the power-weighted centroid. We investigate whether cavity asymmetry improves localisation using microphone amplitude and phase measurements, which would let a sensor monitor beam displacement from its own acoustic signal. A finite-element model of a dumbbell cell predicts that unequal cavities move the microphone ratio away from unity. We printed a symmetric and an asymmetric cell, placed a microphone in each cavity, and translated a compact optically heated absorber along the axis. In the asymmetric cell the ratio at the centre rose to 2.84, close to the predicted 2.82. Within-scan validation recovered the stage-referenced axial position with a mean absolute error of 1.98 mm, compared with 8.65 mm for the symmetric cell, a fourfold difference under the same regression procedure. The results motivate further investigation of cavity geometry for axial localisation.