Tiny bolt-head deformations offer a route toward optical force checks

Looking beyond visible looseness

The force established when a bolt is tightened, known as preload, helps hold joined parts together so that friction can transmit external loads. Insufficient preload can allow the parts to separate earlier and increase the fluctuating stresses carried by the bolt, affecting fatigue behaviour.

Tightening torque is a convenient but indirect guide to this force. Friction in the threads and beneath the bolt head consumes much of the applied torque, and changes in lubrication, coatings or surface condition can change the resulting preload.

In an arXiv preprint posted on November 6, 2026, Nehal Afifi and colleagues propose PreForceVision, a framework for estimating preload from deformation of the exposed bolt-head face. Rather than looking for rotation or an obvious gap, it targets a much smaller mechanical response: changes in surface shape and strain under load.

The distinction is important. This paper reports preliminary evidence that the deformation exists and is measurable with specialist instruments. It does not demonstrate a complete image-based preload prediction system.

What the preliminary measurements show

The reference case was an M16 hex-head bolt of strength class 8.8, with the joint geometry held fixed.

A finite-element simulation at a preload of 90 kilonewtons predicted a shallow, concave deformation of the head face, with a central depression of approximately 12 micrometres. The simulated surface strain was approximately 0.1%.

Separate measurements from three strain gauges on a sensor-integrated bolt also showed a load-dependent surface response. Mean compressive strain was −0.046% at 30 kilonewtons, −0.082% at 60 kilonewtons and −0.116% at 90 kilonewtons. Those values were consistent in order of magnitude with the simulation’s strain result; they were not direct measurements of the central depression.

A third comparison used a Keyence VL-800 optical 3D measuring instrument, which projects light patterns to measure surface shape. Measurements taken before and after assembly at a tightening torque of 200 newton-metres were aligned computationally. The resulting comparison showed a central depression of roughly 10–20 micrometres relative to the edge region.

That optical comparison reproduced the simulation’s broad spatial pattern. However, the paper reports the assembly torque for this measurement, not an independently measured preload establishing that it matched the simulation’s 90-kilonewton condition.

How the proposed image method would work

PreForceVision would compare two images of the same bolt head: one taken before loading or at a known baseline, and another after loading or during monitoring. A single photograph is not the proposed starting point.

The images would first be aligned to correct camera translation, rotation, scale and perspective. Otherwise, camera movement could be mistaken for material deformation. Processing would also reduce lighting differences and reflections.

Digital image correlation—a technique that tracks small patches of surface texture—or related motion-tracking methods would then estimate local displacement. From this, the system would calculate physically interpretable features such as radial strain, centre-to-edge displacement differences and deformation symmetry.

The proposed prediction stage would relate those features to preload and an uncertainty estimate. The authors recommend testing simpler regression models before evaluating a more elaborate neural network combining paired images, deformation maps and known experimental information. These are proposed methods, not reported predictive results.

The gap between laboratory evidence and practical inspection

Detecting a 10–20-micrometre shape change with high-resolution 3D metrology does not establish that ordinary cameras or smartphones can recover a useful force signal. Metallic reflections, weak surface texture, focus, lighting and imperfect alignment could overwhelm it.

Prepared speckle patterns or matte coatings could help image tracking, but would reduce practical convenience. The method also depends on retaining a reference image of the same surface. Performance across different bolts, joint geometries and camera setups cannot be assumed from this fixed-geometry investigation.

The proposed validation programme would use independent force measurements, repeated loading cycles and tests on specimens excluded from training. Keeping specimens separate matters because a model might otherwise learn bolt-specific scratches or markings rather than deformation. Prediction uncertainty would also need testing against actual errors.

The preprint reports no end-to-end preload prediction accuracy or demonstrated smartphone capability. Its contribution is a mechanically supported feasibility case and a structured plan for testing optical estimation—not a replacement for established high-accuracy preload measurements.