This work introduces a novel artificial neural network (ANN)-powered phase field model, offering rapid and precise predictions of fracture propagation in brittle materials. To improve the capabilities of the ANN model, we incorporate a loop of conditions into its core to regulate the absolute percentage error for each observation point, that filters and consistently selects the most accurate outcome. This algorithm enables our model to better adapt to the highly sensitive validation data arising from varying configurations. The effectiveness of the approach is illustrated through three examples involving changes in the microgeometry and material properties of steel fiber-reinforced high-strength concrete structures. Indeed, the predicted outcomes from the improved ANN phase field model in terms of stress–strain relationship, and crack propagation path demonstrates an outperformance compared with that based on the extreme gradient boosting method, a leading regression machine learning technique for tabular data. Additionally, the introduced model exhibits a remarkable speed advantage, being 180 times faster than traditional phase field simulations, and provides results at nearly any fiber location, demonstrating superiority over the phase field model. This study marks a significant advancement in the application of artificial intelligence for accurately predicting crack propagation paths in composite materials, particularly in cases involving the relative positioning of the fiber and initial crack location.

This study proposes a new approach to construct predictive formulas for the permeability of Pervious Concrete (PC), which depends on PC mixture and porosity. To achieve this, a dataset of 195 samples collected from different sources was used. In the dataset the permeability is dependent on porosity, aggregate-to-cement ratio (AC), maximum nominal sizes (MS) of coarse aggregate, and water-to-cement or binder ratios (WC). From the dataset and through applying simple regression techniques, several analytical functions based on the Kozeny-Carman model were constructed and evaluated for their effectiveness in implementing independent datasets and similar analytical functions. Furthermore, for the first time, the Genetic Programming-based Symbolic Regression method was adopted to construct hybrid models combined with the Kozeny-Carman analytical model. The equation of the hybrid model ensures both basic physical conditions and efficiency while being simple enough for engineering-level applications.

In highly filled particulate composites, the particle-size distribution and the resulting packing state limit how much reinforcement a formulation can carry and strongly influence its macroscopic stiffness, yet neither is easily incorporated into predictive micromechanical models. Classical schemes treat inclusions as abstract phases, missing both the maximum-packing (crowding) effect and the influence of particle-size polydispersity on the effective modulus; existing Morphologically Representative Pattern (MRP)-based extensions that do incorporate a finite packing fraction remain restricted to a single particle size. This paper extends the two-pattern MRP framework of [Marcadon, Hervé and Zaoui (2007)] to particle-size polydispersity by introducing a second composite pattern with a size-ratio parameter γ that captures the geometrical interaction between differently sized particles, linking measurable descriptors of the particle-size distribution and packing state directly to the effective stiffness within a single analytical formulation. The resulting scheme reduces exactly to the monodisperse MRP model when γ = 0, and to the Einstein dilute estimate as the particle volume fraction tends to zero. For bidisperse rigid particles in a yield-stress fluid, it reduces the linear-space mean squared error against experimental data from 0.0613 (best-fit empirical power law) to 0.0316; for bidisperse diamond particles in a metal matrix it achieves a mean squared error of 0.0014 against a small dataset, with predicted stiffness comparatively robust to the closure relation's calibration constants. The formulation thus provides an analytical link between particle-size distribution, packing state and effective stiffness, evaluable before more detailed microstructural simulations.

A monitoring system records one signal and must attribute it to two unknowns: the traffic that crossed and the structure that carried it. This paper asks when the two can be told apart from response-integral measurements, treating it as a question of observability. Linearizing about a reference state gives load- and deterioration-sensitivity matrices whose ranges are two subspaces, and the principal angles between them measure separability. On a two-span deck of thirteen girder lines with six channels the joint rank condition holds, so the causes are locally identifiable within the linearized model; but two of three principal angles fall below 2°, the separability index is 0.013 and the condition number is 432, so load and deterioration are nearly confounded. The near-alignment is mechanical: a load increment and a stiffness decrement reach the measurement through the same inverse operator. Two measurement directions escape, and they are operational duals. The first is a conservation law: the width-summed kernel of any admissible linear measurement equals that sensor’s reading under a uniform load, and is invariant to transverse-rigidity changes within the stated model class, its leakage from transverse deterioration being 113 times smaller than from uniform loss. The invariance relies on two kinematic hypotheses, aligned supports and free longitudinal edges, both tested through failure cases, the second identified by an independent analytic plate cross-check. The dual direction is null to load magnitude, with an exactly characterized mode in which its sensitivity vanishes. Stacking crossings shows that traffic diversity adds deterioration rank only where instrumentation is short of channels.