luviner/industrial-faults

--- license: cc-by-4.0 task_categories: - tabular-classification - time-series-classification tags: - anomaly-detection - predictive-maintenance - synthetic-data - industrial - fault-detection - bearing-fault - time-series pretty_name: Luviner Industrial Fault Dataset size_categories: - 1K<n<10K language: - en --- # Luviner Industrial Fault Dataset **6,500 labeled samples** across **13 classes** (1 normal + 12 industrial fault types) with 8 sensor features. Generated by the [Luviner AI](https://luviner.com) synthetic anomaly engine — 12 parametric failure mode generators that produce temporally progressive, physically realistic fault signatures. ## Dataset Description ### Why synthetic industrial data? Real industrial failure data is extremely scarce — machines rarely fail, and when they do, the data is often proprietary. This dataset provides realistic labeled fault data for: - **Training** anomaly detectors and fault classifiers - **Benchmarking** time-series classification models - **Education** on industrial fault signatures - **Prototyping** predictive maintenance systems ### Features (8 sensor channels) | Feature | Description | |---------|-------------| | `vibration_x` | Vibration amplitude, X-axis | | `vibration_y` | Vibration amplitude, Y-axis | | `vibration_z` | Vibration amplitude, Z-axis | | `temperature` | Surface temperature | | `pressure` | System pressure | | `current` | Motor current draw | | `flow_rate` | Process flow rate | | `acoustic_db` | Acoustic emission level | ### Classes (13) | Label | Fault Type | Category | |-------|-----------|----------| | 0 | normal | — | | 1 | bearing_degradation | mechanical | | 2 | shaft_imbalance | mechanical | | 3 | gear_tooth_crack | mechanical | | 4 | misalignment | mechanical | | 5 | thermal_runaway | thermal | | 6 | thermal_cycling | thermal | | 7 | electrical_fault | electrical | | 8 | sensor_drift | electrical | | 9 | intermittent_contact | electrical | | 10 | cavitation | process | | 11 | blockage | process | | 12 | leakage | process | ### Temporal progression Each fault type has **temporally progressive** characteristics — faults evolve from subtle to severe over the sample sequence. This is realistic: industrial failures don't appear from nothing, they develop over time. ## Files | File | Samples | Description | |------|---------|-------------| | `luviner_industrial_faults.csv` | 6,500 | Main dataset (500/class, severity 1.0) | | `luviner_faults_mild.csv` | 2,600 | Mild faults (200/class, severity 0.5) | | `luviner_faults_moderate.csv` | 2,600 | Moderate faults (200/class, severity 1.5) | | `luviner_faults_severe.csv` | 2,600 | Severe faults (200/class, severity 2.0) | ## Quick Start ```python import pandas as pd from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestClassifier from sklearn.metrics import classification_report df = pd.read_csv("luviner_industrial_faults.csv") X = df.drop(columns=["label", "fault_type"]) y = df["label"] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) clf = RandomForestClassifier(n_estimators=100, random_state=42) clf.fit(X_train, y_train) print(classification_report(y_test, y_test, target_names=df["fault_type"].unique())) ``` ## Generate your own Want more data, different features, or custom failure modes? Use the [Luviner API](https://luviner.com/en/pricing/api): ```bash curl -X POST https://luviner.com/api/v1/synthetic/generate \ -H "Authorization: Bearer YOUR_TOKEN" \ -H "Content-Type: application/json" \ -d '{"failure_mode": "bearing_degradation", "n_features": 8, "n_samples": 1000, "severity": 1.5}' ``` Also available on [RapidAPI](https://rapidapi.com/luviner/api/luviner). ## Citation ```bibtex @dataset{luviner_industrial_faults_2026, title={Luviner Industrial Fault Dataset}, author={Luviner Edge AI}, year={2026}, url={https://huggingface.co/datasets/luviner/industrial-faults}, note={Synthetic industrial failure data from 12 parametric fault generators} } ``` ## License CC-BY-4.0 — free to use, share, and adapt with attribution.--- license: CC BY 4.0（知识共享署名4.0国际许可协议） task_categories: - 表格分类（tabular-classification） - 时间序列分类（time-series-classification） tags: - 异常检测（anomaly-detection） - 预测性维护（predictive-maintenance） - 合成数据（synthetic-data） - 工业领域（industrial） - 故障检测（fault-detection） - 轴承故障（bearing-fault） - 时间序列（time-series） pretty_name: Luviner工业故障数据集 size_categories: - 1000 < 样本数 < 10000 language: - 英语 --- # Luviner工业故障数据集 **6500条带标注样本**，覆盖**13个类别**（1个正常类别与12种工业故障类型），共包含8项传感器特征。 本数据集由[Luviner AI](https://luviner.com)的合成异常引擎生成——依托12种参数化故障模式生成器，生成具备时间渐进特性、符合物理真实度的故障特征信号。 ## 数据集说明 ### 为何选用合成工业数据？ 真实工业故障数据极为稀缺——工业设备极少发生故障，即便发生故障，相关数据也多为专有资产。本数据集可为以下场景提供具备真实可信度的带标注故障数据： - **训练**异常检测器与故障分类模型 - **基准测试**时间序列分类模型性能 - **教学**：开展工业故障特征相关的专业教育 - **原型开发**：搭建预测性维护系统原型 ### 8项传感器通道特征 | 特征名称 | 特征描述 | |---------|-------------| | `vibration_x` | X轴振动幅值 | | `vibration_y` | Y轴振动幅值 | | `vibration_z` | Z轴振动幅值 | | `temperature` | 设备表面温度 | | `pressure` | 系统压力 | | `current` | 电机工作电流 | | `flow_rate` | 工艺流程流量 | | `acoustic_db` | 声发射分贝值 | ### 13个故障类别 | 标签编号 | 故障类型 | 类别归属 | |-------|-----------|----------| | 0 | 正常工况 | — | | 1 | 轴承退化 | 机械类 | | 2 | 轴不平衡 | 机械类 | | 3 | 齿轮齿裂纹 | 机械类 | | 4 | 装配不对中 | 机械类 | | 5 | 热失控 | 热学类 | | 6 | 热循环 | 热学类 | | 7 | 电气故障 | 电气类 | | 8 | 传感器漂移 | 电气类 | | 9 | 接触不良 | 电气类 | | 10 | 气蚀 | 工艺类 | | 11 | 管路堵塞 | 工艺类 | | 12 | 介质泄漏 | 工艺类 | ### 时间渐进特性 每种故障类型均具备**时间渐进特性**——在样本序列中，故障会从轻微症状逐步发展为严重故障。这符合真实工业场景：工业故障不会凭空出现，而是随时间推移逐步演化。 ## 数据集文件 | 文件名称 | 样本数量 | 文件说明 | |------|---------|-------------| | `luviner_industrial_faults.csv` | 6,500 | 主数据集（每类别500条样本，严重程度为1.0） | | `luviner_faults_mild.csv` | 2,600 | 轻度故障数据集（每类别200条样本，严重程度为0.5） | | `luviner_faults_moderate.csv` | 2,600 | 中度故障数据集（每类别200条样本，严重程度为1.5） | | `luviner_faults_severe.csv` | 2,600 | 重度故障数据集（每类别200条样本，严重程度为2.0） | ## 快速上手 python import pandas as pd from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestClassifier from sklearn.metrics import classification_report df = pd.read_csv("luviner_industrial_faults.csv") X = df.drop(columns=["label", "fault_type"]) y = df["label"] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) clf = RandomForestClassifier(n_estimators=100, random_state=42) clf.fit(X_train, y_train) print(classification_report(y_test, y_test, target_names=df["fault_type"].unique())) ## 自定义生成数据 如需获取更多数据、更换特征维度或自定义故障模式，可使用[Luviner API](https://luviner.com/en/pricing/api)： bash curl -X POST https://luviner.com/api/v1/synthetic/generate -H "Authorization: Bearer YOUR_TOKEN" -H "Content-Type: application/json" -d '{"failure_mode": "bearing_degradation", "n_features": 8, "n_samples": 1000, "severity": 1.5}' 本数据集也可在[RapidAPI平台](https://rapidapi.com/luviner/api/luviner)获取。 ## 引用格式 bibtex @dataset{luviner_industrial_faults_2026, title={Luviner Industrial Fault Dataset}, author={Luviner Edge AI}, year={2026}, url={https://huggingface.co/datasets/luviner/industrial-faults}, note={Synthetic industrial failure data from 12 parametric fault generators} } ## 许可协议 采用CC BY 4.0许可协议——可自由使用、分享与改编本数据集，但需注明原作者。