Overview
This dataset contains a synthetic benchmark for studying data-driven assessment of fresh concrete consistency from mixer power consumption time series ("mixing curves"). It was created in the context of the manuscript
Learning from Mixing Power Curves: A Long Short-Term Memory-Based Soft Sensor for Online Assessment of Fresh Concrete Consistency (submitted to the Journal of Building Engineering).
The data links randomized concrete mix designs, simulated mixing power curves, and simulated slump flow values as a consistency measure. It is intended for developing, pretraining, and evaluating machine learning models that use process data from concrete mixing plants, with a focus on long short-term memory (LSTM) networks and other sequence models.
Data contents
All tabular data are provided in machine-readable formats (e.g. CSV) with one row per mix and clearly named columns. The dataset consists of:
Mix design parameters
Cement, filler, aggregate and water contents per m³, water-to-powder ratio, and related scalar descriptors used in the simulation.
Mixing power curves
Synthetic mixing curves with a fixed sequence length (90 time steps). Each curve represents the electrical power consumption of a single (hypothetical) mixer over one batch, simulated according to an empirically supported six-stage mixing model.
Reference slump flow values
Simulated slump flow for each mix, derived from empirical relationships between mix proportions, rheological parameters and slump flow reported in the literature.
Simulation framework
Mixing curves are generated using a six-stage concrete mixing kinetics model that captures key transitions such as granule formation, fluidity point and subsequent stabilization of power consumption. Slump flow labels are computed from the mix design and late-stage mixing power values using empirical relationships between mix proportions, rheological parameters, and slump flow.
The dataset therefore offers a controlled environment to test whether and how sequence models can learn to infer fresh concrete consistency from the temporal development of mixing power, beyond what can be obtained from mix proportions and single end-of-mixing power values.
Intended use
The dataset is designed for:
- Pretraining and benchmarking sequence models (e.g. LSTM, GRU, transformers)
- Comparing sequence-based and static models for consistency prediction
- Method development for online quality assessment in concrete production
- Teaching and demonstration of AI methods in civil / materials engineering
Variable Definition
- torque_x: A series of 90 time-steps reflecting the mixer's power consumption from start until end of mixing. One value per time-step.
- water, cement, filler, aggregate: (content of given mix constituents)
- slump flow value: The slump flow value based on the mix proportion as well as the power consumption in mixing stage VI
Variable definitions, units and simulation parameters are documented in greater detail in the related publication.
Version history
- 2025-03: Initial upload.
- 2025-12: Updated version of the dataset including minor fixes and an updated terminology / naming convention across the board.
(2025-12-12)
