PICUS 2

What Comes Next:

PICUS 2.0, or "How to Make Every Tap a Measurement"

Percussion testing has resisted quantification for as long as it has existed, and the reason is subtle. What the instrument records is the sound of the tap — but that sound is the response of the surface convolved with an excitation that changes from strike to strike and from operator to operator. Any index computed on it therefore describes the object and the way it was struck at the same time, with no way to separate the two.

The next phase of our research removes that confusion at the root. Since the instrument already measures the impact force at every strike, each percussion can be treated as a complete input–output identification experiment: the force measured by an accelerometer rigidly coupled to the striker and the sound pressure measured by the microphone are processed together, and what is recovered by regularized deconvolution is the impulse response of the struck surface — a property of the surface alone. Every diagnostic index is then computed on that response rather than on the raw sound.

What this changes

Independence from the strike. Variations in impact strength, striker rebound and actuator repeatability are compensated automatically, because the input is measured rather than assumed. The same point can be re-measured months later, by a different operator striking with a different force, and the results remain comparable.
A built-in quality figure. The same averaged spectra that yield the response estimate also yield the coherence function, which quantifies — frequency by frequency and point by point — how much of the recorded sound is linearly explained by the measured force. Measurement quality stops being an act of faith and becomes a number the operator can read.
Strikes spent where they are needed. Because coherence also governs the estimation uncertainty, the instrument can decide how many times to strike each point: well-coupled points on sound surfaces are dispatched in a few taps, while noisy or poorly coupled ones automatically receive more, up to a hard limit.
Physical geometry instead of scores. Each strike carries two independent observables. The contact time, resolved on a fast force channel, inverts to the thickness of the vibrating layer; the resonance frequency of the estimated response, combined with that thickness, inverts a calibrated thin-plate model to give the equivalent lateral radius of the detachment. The output becomes two dimensions rather than an arbitrary index.
Continuity with contactless imaging. The reflection-index and absorption-matrix formalism developed for contactless vibro-acoustic imaging transfers unchanged, because it was already written on the impulse response — so the hand-held instrument produces maps directly comparable with those of the contactless technique, at a fraction of the hardware complexity.

Alongside these, the response yields a set of physically interpretable indices for every mapped point: resonance frequency and quality factor, decay rate obtained from backward integration of the response, spectral centroid, and band-integrated reflection indices. A bonded surface radiates a damped, low-Q response and dissipates energy into the substrate; a detachment rings.

The instrument

PICUS 2.0 keeps the field-proven mechanical concept of its predecessor — a hand-held probe with a solenoid-driven percussor, infrared positioning and battery operation — and rebuilds the sensing chain around the identification principle. The probe carries a wide-band MEMS microphone and a high-g, DC-coupled accelerometer fixed to the striker, conditioned by two paths with deliberately different goals: the acoustic path band-limited for the spectral estimate, the force path kept DC-coupled and sampled far faster, because the contact-time information lives in the fidelity of the pulse edges. A floating-point microcontroller performs the windowing, transforms, spectral accumulation and index extraction on line, so the instrument stays standalone in the field.

What the architecture does not require is worth stating as well: no anechoic conditions, no calibrated excitation, no contact with the surface beyond the strike itself, and no operator skill beyond positioning the probe.

The planned programme

The theoretical framework and its numerical validation against synthetic references are established; the experimental work will proceed in four stages, each answering one question.

Bench validation. Does the hardware chain reproduce, on real percussions, the estimator behaviour certified in simulation? The test is deliberately falsifiable: indices computed on the raw sound must vary with strike strength, indices computed on the deconvolved response must not.
Calibration on reference specimens. Specimens with engineered circular detachments of known radius and depth, under plaster and tile layers of graded thickness, will calibrate both geometric channels and establish detection thresholds — the smallest detachment radius and the greatest depth that can still be separated from the bonded baseline with a stated confidence.
Cross-validation against the contactless technique. On shared specimens and at least one real surface, the maps produced by the two methods will be compared directly. Agreement would establish them as interchangeable front ends of a single imaging framework: contactless where no contact at all is admissible, percussive where portability, cost and speed dominate.
Field campaigns. Supervised campaigns on frescoed and tiled surfaces of documented conservation history, and on civil substrates such as rendered masonry and clad façades, will assess scan throughput, repeatability across operators and sessions, and the diagnostic value of the geometric outputs as judged by conservators and structural engineers.

Why it matters

For conservation, the step from a qualitative score to a physical geometry is operationally concrete: an estimate of the thickness and lateral extent of a detachment, with a per-point confidence figure, feeds directly into planning a consolidation campaign, dosing the consolidants and monitoring a known defect over time.

Beyond heritage, the same mechanical scheme — a stiff finishing layer imperfectly bonded to a massive substrate — describes a large part of the built environment: render on masonry, tiled and stone-clad façades whose spontaneous detachment is a recognised public-safety problem in dense cities, screeds and repair overlays on concrete decks. There the instrument occupies a niche complementary to established impact-echo practice: airborne on the receiving side, far cheaper and faster than laser vibrometry, and, unlike a traditional sounding survey, producing documented, georeferenced, operator-independent maps with an explicit uncertainty channel.

Finally, what each point returns is not a single number but a structured vector — resonance, damping, decay, centroid, reflection indices, geometric estimates and coherence. Vectors of this kind are natural inputs for statistical classification, and the coherence supplies exactly the sample weighting that such training pipelines usually lack. That is the line of work the campaigns described above are also designed to seed.

This page describes research in progress. Results will be reported here as the experimental programme advances.