Dataset card for FactoryNet v4: 56,591 episodes, 7 embodiments, per-subset licenses

README.md

@@ -1,74 +1,177 @@
 ---
+license: other
+license_name: per-subset
+license_link: https://huggingface.co/datasets/Forgis/FactoryNet#licensing
+pretty_name: FactoryNet
+size_categories:
+  - 100M<n<1B
+task_categories:
+  - time-series-forecasting
+  - tabular-classification
+tags:
+  - industrial
+  - robotics
+  - anomaly-detection
+  - time-series
+  - sim-to-real
+  - manufacturing
+  - predictive-maintenance
 configs:
-  - config_name: raw
+  - config_name: default
     data_files:
       - split: train
-        path: data/raw/*.parquet
-  - config_name: normalized
-    data_files:
-      - split: train
-        path: data/normalized/*.parquet
----
-
-# 🏭 FactoryNet: A Unified Multi-Machine Industrial Dataset
-
-## Overview
-FactoryNet is a large-scale, machine-learning-ready foundation dataset for industrial robotics and manufacturing anomaly detection. Historically, industrial datasets have been heavily siloed—every manufacturer and research team uses different column names, units, and structures. 
-
-FactoryNet solves this by forging massive, high-frequency physical datasets from completely different machines into a **single, mathematically unified coordinate system**. By standardizing axes, effort signals, and kinematic feedback, this dataset allows neural networks to learn universal physical relationships across different hardware boundaries.
-
-## 📦 Dataset Composition
-This repository currently contains millions of rows of high-frequency sensor data merged from three distinct open-source industrial datasets:
-
-### 1. UMich CNC Mill Tool Wear Dataset
-* **Machine:** 3-Axis CNC Mill
-* **Task:** Machining wax blocks under varying feedrates and clamp pressures.
-* **Anomalies:** Tool wear (Unworn vs. Worn) and visual inspection failures.
-* **Original Source:** University of Michigan (via Kaggle)
-
-### 2. AURSAD (Automated UR3e Screwdriving Anomaly Dataset)
-* **Machine:** UR3e 6-Axis Collaborative Robot
-* **Task:** Automated screwdriving using an OnRobot Screwdriver.
-* **Anomalies:** Normal operation, damaged screws, missing screws, extra parts, and damaged threads.
-* **Original Source:** Zenodo (Record 4487073)
-
-### 3. voraus-AD (Yu-Cobot Pick-and-Place)
-* **Machine:** Yu-Cobot 6-Axis Collaborative Robot
-* **Task:** Industrial pick-and-place task on a conveyor belt.
-* **Anomalies:** 12 diverse physical anomalies including axis wear (friction/miscommutation), gripping errors, collisions, and added axis weights.
-* **Original Source:** voraus robotik (via Kaggle)
-
+        path: data/*.parquet
 ---
 
-## 🏗️ The Unified Schema
-To allow cross-machine learning, all raw variables (which previously had over 300 conflicting names) have been mapped to a standardized `FactoryNet` schema. 
-
-**Standardized Prefix Naming:**
-* `setpoint_*`: The commanded target from the controller (e.g., `setpoint_pos_0`).
-* `feedback_*`: The actual measured state from the sensors (e.g., `feedback_vel_1`).
-* `effort_*`: The physical force/current applied (e.g., `effort_current_2`, `effort_torque_0`).
-* `ctx_*`: Contextual metadata (e.g., `ctx_anomaly_label`, `ctx_busvoltage_0`).
+# FactoryNet
 
-**Standardized Axis Indexing:**
-Regardless of how the original manufacturer numbered their joints (X/Y/Z or 1-6), all axes in this dataset are strictly zero-indexed (`0` through `5`).
+A multi-embodiment industrial time-series corpus: **56,591 end-to-end task executions**
+(14,847 real, 41,744 simulated), **113M logged timesteps**, **7 embodiments**, **4 tasks**,
+**27 annotated anomaly types** with healthy baselines and counterfactual pairs.
 
----
-
-## ⚙️ Configurations
-This dataset is partitioned into highly compressed Parquet files and is available in two configurations:
+Every signal from every source is mapped into one control-theoretic schema —
+**Setpoint, Effort, Feedback, Context (S-E-F-C)** — so a single dataloader works across a
+6-DOF arm and a 4-axis CNC gantry, and *commanded versus realized* dynamics are readable
+as explicit prediction residuals rather than opaque reconstruction scores.
 
-1. **`raw`**: The original physical values (Amps, Volts, Radians, etc.) mapped directly into the new schema. Best for physics-informed neural networks or domain-specific thresholding.
-2. **`normalized`**: All continuous physical variables have been independently standardized using a Z-score Scaler (`StandardScaler`) fitted specifically to that machine's domain. Best for immediate deep learning and foundation model training.
 
-## 🚀 How to Use (Python)
-Because this dataset is partitioned using Parquet, you can load the entire multi-gigabyte repository without crashing your local RAM.
+## Quick start
 
 ```python
 from datasets import load_dataset
 
-# Load the mathematically normalized dataset for AI training
-dataset = load_dataset("karimm6/FactoryNet_Dataset", "normalized")
+ds = load_dataset("Forgis/FactoryNet", split="train", streaming=True)
+print(next(iter(ds)))
+```
 
-# Convert to a Pandas DataFrame
-df = dataset['train'].to_pandas()
-print(df['machine_type'].value_counts())
+Or read the Parquet directly, which is usually what you want for time series.
+`cnc_000.parquet` is 1.1 MB, so this is a fast first look:
+
+```python
+import pandas as pd
+from huggingface_hub import hf_hub_download
+
+p = hf_hub_download("Forgis/FactoryNet", "data/cnc_000.parquet",
+                    repo_type="dataset")
+df = pd.read_parquet(p)
+
+# the same four prefixes on every machine, without vendor-specific names
+S = [c for c in df.columns if c.startswith("setpoint_")]
+E = [c for c in df.columns if c.startswith("effort_")]
+F = [c for c in df.columns if c.startswith("feedback_")]
+C = [c for c in df.columns if c.startswith("ctx_")]
+```
+
+The same five lines work unchanged on a 6-DOF arm. The consolidated real UR
+recording, `data/factorywave_ur_consolidated.parquet`, is 992 MB with 7,129,261
+rows and 125 columns (44 Setpoint, 19 Effort, 25 Feedback, 30 Context), so pass
+`columns=[...]` to `read_parquet` rather than loading it whole.
+
+### What is where
+
+| Prefix in `data/` | Source | Files |
+|---|---|---|
+| `factorywave_*` | our laboratory UR3 and KUKA KR10 recordings, incl. screwdriver telemetry and episode metadata | 6 |
+| `voraus_*` | voraus-AD | 60 |
+| `aursad_*` | AURSAD | 32 |
+| `cnc_*` | UMich CNC milling | 1 |
+| `simulations_baseline_*` | simulated nominal episodes | 11 |
+| `simulations_counterfactual_*` | matched simulated counterfactuals | 10 |
+
+`data/` is 13.7 GB of Parquet. The repository also carries the paired sim-to-real
+gap artifacts used for the validation campaign: `real_csv/` and `sim_csv/` (episodes
+paired by filename), `pick_configs/` (per-episode simulation parameters), and
+`gap_reports/` + `summary/` (per-episode and aggregated gap analysis).
+
+## The S-E-F-C schema
+
+One wide table per source: one row per control tick, joined to an episode table by
+`episode_id`. The column prefix carries the role; the suffix `_i` is the positional joint
+index 0..5 (base → wrist), so the same column means the same slot on every arm. A column a
+source does not expose is written as null rather than dropped, so all sources share one
+schema.
+
+| Role | Prefix | What it is | Examples |
+|---|---|---|---|
+| **Setpoint** | `setpoint_*` | commanded intent | target joint position, velocity, acceleration; target TCP pose; `gripper_command` |
+| **Effort** | `effort_*` | actuation energy expended | motor current, joint torque, commanded torque, estimated contact force |
+| **Feedback** | `feedback_*` | measured physical outcome | encoder position and velocity, TCP pose, tool accelerometer |
+| **Context** | `ctx_*` | environment and static state | payload mass, task phase, fault labels, safety/robot mode, speed scaling, I/O bits |
+
+This separation is the point of the corpus. Existing industrial datasets log sensor
+*outcomes* without distinguishing what the controller asked for from what the machine did,
+which makes cross-machine dynamics learning hard to set up at all.
+
+## Composition (the corpus as reported in the paper)
+
+| Source | Machine | Tasks | Faults | Episodes | Timesteps |
+|---|---|---|---|---|---|
+| Lab (real) | UR3 | P&P, Screw, Peg | yes | 8,863 | 11M |
+| Lab (real) | KUKA KR10 | P&P | yes | 1,799 | 4M |
+| Open (real) | voraus-AD (Yu-Cobot) | P&P | yes | 2,122 | 16M |
+| Open (real) | AURSAD (UR3e) | Screw | yes | 2,045 | 3M |
+| Open (real) | UMich CNC | Machining | yes | 18 | 18K |
+| Synthetic | Isaac Sim — UR3, UR5, UR10, UR30, KR10 | P&P | yes | 41,744 | 79M |
+| **Total** | | | | **56,591** | **113M** |
+
+1,553 real counterfactual episodes accompany the faulty runs. Of the 10,662 lab episodes,
+~28% are healthy and ~72% contain an injected fault.
+
+### Synthetic track — read this before comparing against the paper
+
+**This snapshot carries the earlier Isaac Sim 4.5.0 synthetic campaign**
+(`simulations_baseline_*` and `simulations_counterfactual_*`), not the five-arm
+41,744-episode campaign the paper reports. That campaign is Isaac Sim 5.1 / Isaac Lab
+2.3 across UR3, UR5, UR10, UR30 and KR10, and is being uploaded during the review
+period. Every real subset here is final and matches the paper.
+
+The 5.1 campaign runs GPU-batched PhysX at 500 Hz, with control and logging at 125 Hz
+for the UR arms and 83.3 Hz for the KR10 (the KRC interpolation cycle), no resampling.
+Six fault classes: payload addition, motor miscommutation, gripper activation failure,
+gripper release, collision, path obstacle.
+
+Every episode passes a two-stage validity gate — structural (plausible length, complete
+phases, part placed, injected event present, non-zero contact force) and trace (every
+fault must leave a measurable physical effect). **41,744 of 50,000 generated episodes
+pass.** This guarantees that a fault label corresponds to a real physical effect, but it
+also biases the synthetic faults toward more detectable instances, and acceptance rates
+differ by fault class. UR3 and KR10 motion profiles are fitted to real recordings; UR5,
+UR10 and UR30 have no real counterpart and are scaled from the UR3 fit.
+
+## Known limitations
+
+Stated here because they determine what the corpus can be used for:
+
+- **UR5, UR10 and UR30 exist only in simulation.** Any multi-arm result on them is sim-to-sim.
+- **The validity gate favours detectable faults**, so synthetic fault difficulty is not representative.
+- **Recording context partly predicts fault labels** on both real robots (program, day, speed override), so real fault evaluation should stay within a single recording session.
+- **Simulated motor current is not calibrated** to the real robots out of the box. Task and cycle structure transfer from simulation; fault signatures largely do not.
+- The KUKA KR10 (KSS 8.3) does not expose joint velocities, commanded TCP pose, or TCP force/torque over RSI; those channels are null.
+
+## Licensing
+
+Licensing is per subset:
+
+| Subset | License |
+|---|---|
+| Laboratory (`factorywave_*`) and synthetic (`simulations_*`) | MIT |
+| `voraus_*` (voraus-AD) | CC BY-NC-SA 4.0: **NonCommercial and ShareAlike**; works derived from it must keep this license |
+| `aursad_*` (AURSAD) | CC BY 4.0 |
+| `cnc_*` (UMich CNC) | follows its source terms |
+
+Cite the original sources when using the adapted subsets.
+
+## Code
+
+Adapters, simulation configs and the transfer study: https://github.com/Forgis-Labs/FactoryNet
+
+## Citation
+
+```bibtex
+@article{petersen2026factorynet,
+  title   = {FactoryNet: A Large-Scale Dataset toward Industrial Time-Series Foundation Models},
+  author  = {Petersen, Jonas and Jawahar, Pratik and Maggioni, Riccardo and Huou, Alexandre and Othman, Karim and Mazzoleni, Camilla and Cozzi, Matteo and Gajewski, Marek and Paloschi, Davide and Lombardi, Gian-Alessandro and Ignuta-Ciuncanu, Matei C. and Porath, Lior and Martelli, Federico and Petersen, Philipp Christian},
+  journal = {arXiv preprint arXiv:2605.09081},
+  year    = {2026}
+}
+```