The stakes of this ambiguity are enormous. In field operations, hard rock and bit degradation can produce an identical macroscopic signature: increasing torque on bit, decreasing rate of penetration, and stronger vibration. A driller crossing into a harder layer may misclassify the change as bit failure, while a worn bit that briefly unloads after entering a softer layer may appear to have recovered. These bidirectional misreadings affect weight-on-bit and rotary-speed selection, the timing of trips to replace bits, estimates of remaining bit life, and the triggering of downhole failure alarms. Until now, most laboratory and numerical studies held either lithology or wear fixed, making it impossible to determine whether the torque increment caused by wear was constant across different rock types.

The missing link, the authors argue, is mechanical rather than merely statistical. Wear changes the effective depth of cut, the frictional contact area, the duration of contact, and the set of cutters actively engaged with the rock. Lithology changes the cutting and indentation coefficients and the manner in which contacts are renewed by fracture. Because full-bit torque is the sum of each cutter’s tangential force multiplied by its radial moment arm, a small change in the shoulder or gauge region of the bit can be strongly amplified. The same nominal wear state can therefore generate different, or even oppositely signed, departures from a simple additive torque model depending on which rock is being drilled.

To isolate this effect, the team built a full-bit model in PFC3D, a discrete-element simulation environment in which rock is represented as an assembly of bonded particles. The model contained 230 rigid wall components and more than 221,000 triangular facets, with weight-on-bit and torque summed consistently from the contact forces and moments of every component. Four controlled lithology proxies were constructed, spanning fine clastic, conglomeratic, pyroclastic, and rhyolitic rocks with densities from 2,450 to 2,650 kilograms per cubic meter and particle-scale stiffnesses from 8 to 16 gigapascals. Wear was imposed as axial edge recession of the cutters at four levels: 0, 50, 100, and 150 micrometers, with cutter friction changed synchronously.

The experimental design was unusually disciplined. An early discovery matrix of 48 cases and a replication matrix of 80 cases were used to identify the interaction pattern and lock the analysis protocol before any response data from the confirmation stage were inspected. Then a third, completely independent matrix was generated: four lithologies, four wear levels, and 12 new random seeds per cell, totaling 192 cases. Every case passed preregistered quality gates covering contact fraction, force-ratio equilibrium, and load reasonableness, and no design point was deleted or replaced according to its response. The result was unambiguous. The full lithology-by-wear interaction on the dimensionless torsional response, defined as torque divided by the product of weight-on-bit and bit radius, yielded F(9, 176) = 5.562, a permutation p-value below 0.0001, and a partial eta-squared of 0.221.

Sensitivity analyses confirmed the finding was not an artifact. Repeating the interaction test with raw torque, torque adjusted for weight-on-bit, and after excluding the lowest 5 percent of weight-on-bit cases all produced consistent results. A separate particle-size sensitivity study across three radius ranges showed the interaction pattern was robust to numerical resolution, and a paired endpoint decomposition of 48 cases revealed that at severe wear the combined response followed mainly the geometric recession term, with the friction contribution averaging nearly zero and its direction varying unpredictably with lithology.

What makes the study remarkable is that it did not stop at detecting the interaction; it localized the mechanism. Using an exact component-wise load identity, the team decomposed the dimensionless torque into three factors: the moment-arm-weighted tangential-force demand, the signed coherence of force directions among components, and the closure between component loads and wall-reaction torque. The interaction was concentrated almost entirely in the first factor, with an effect size matching that of the overall response, while the closure factor showed no interaction at all. This means the nonadditivity arose because lithology-dependent fracture and wear-dependent engagement changed which cutters carried tangential force and at what radius, not from any numerical artifact or simple cancellation of forces.

The practical consequences were quantified through a mechanism-conditioned interaction compensator, a ridge-regression predictor trained on observable quantities such as tangential-force direction, force ratios, and moment-weighted radii. In complete-cell holdout tests, where entire lithology-wear combinations were excluded from fitting, adding force direction raised explanatory power from 3.0 percent to 67.2 percent, and adding active-contact topology pushed it to 84.0 percent. The compensator reduced the interaction sum of squares by 78.0 percent and cut the maximum cellwise bias from 32.7 percent to 13.4 percent. At the same moderate wear level, fine clastic rock showed a torque deviation of minus 24.8 percent from the additive expectation while rhyolitic rock showed plus 32.7 percent, a stark demonstration that a universal wear correction or fixed threshold will fail in one direction or the other.

Field data from 29 complete drilling runs in a heterogeneous volcanic-sedimentary sequence reinforced the numerical conclusions. Meter-scale drilling-time trajectories captured both fast responses at lithologic boundaries and slow within-run drift, with each additional 100 meters of drilling multiplying drilling time by 1.205. Severe bit damage occurred far more often in the harder lithologies, with a pooled odds ratio of 37.5, yet the agreement between severe physical damage and performance-related pull decisions was only 60.7 percent, showing that operational labels are an unreliable proxy for actual bit condition. The authors propose deploying their compensator as a bypass correction layer alongside existing top-drive torque monitoring: the raw torque limit continues to protect equipment, while the corrected value identifies abnormal torsional demand relative to the current lithology and contact state. By subtracting the state-dependent nonadditive offset before applying any alarm threshold, the approach promises to reduce false hard-rock and wear alarms, improve detection of genuine degradation, and supply an interpretable correction term for drilling digital twins.

Subject of Research: Nonadditive coupling between PDC drill bit wear and rock lithology in full-bit torsional response

Article Title: Modulation mechanism of full-bit torsional response under nonadditive coupling between PDC-bit wear and lithology

Article References: Mei, D., Yan, L., Zhou, Z., Wu, X., & Jian, Y. (2026). Modulation mechanism of full-bit torsional response under nonadditive coupling between PDC-bit wear and lithology. Results in Engineering, 32, Article 113176. https://doi.org/10.1016/j.rineng.2026.113176

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113176

Keywords: PDC bit, drilling torque, bit wear, lithology, nonadditive interaction, PFC3D simulation, torsional response, rock cutting, drilling automation, moment arm, contact topology, drilling digital twin