BRPHM-datasets

# BRPHM Gold-Standard Remaining Useful Life Dataset Collection > **Release status / 发布状态:** 达到公开发布标准 > **Release form / 发布形态:** immutable raw records + standardized unit-level Parquet + machine-verifiable lineage > **Language / 语言:** Chinese and English ## Dataset Card / 数据集卡 | Field / 字段 | Value / 内容 | | --- | --- | | Purpose / 目标 | Remaining useful life research, reproducible data engineering, and source-aware evaluation. / 剩余寿命研究、可复现数据工程与来源感知评测。 | | Unit of release / 发布单元 | Raw source artifacts and normalized unit time series. / 原始来源工件与规范化单元级时间序列。 | | Canonical normalized format / 规范格式 | Apache Parquet, one `unit_*.parquet` file per unit; partition `meta.json` defines fields and units. / 每个单元一个 Parquet 文件，分区 `meta.json` 定义字段与单位。 | | Public authorization / 公开授权 | The publisher declares complete public-release authorization. Supporting records are archived before upload in `certificates_to_attach/`; that directory is not a machine gate. / 提供方声明公开授权完备；证明材料上传前归档，且该目录不参与机器门禁。 | | Excluded material / 排除内容 | Held-out evaluation data, processed training tensors, models, experiment results, `raw/sample_dataset`, and hidden alternate simulation work. / 封存评测数据、训练张量、模型、实验结果、混合样本目录与隐藏替代仿真目录。 | ## Choose A Tier / 三档选择 | Tier / 档位 | Files / 文件数 | Size / 大小 | Recommended use / 推荐用途 | | --- | ---: | ---: | --- | | `BRPHM_RUL_mini` | 8 | 3,112,496 bytes | Schema smoke tests / 格式联调 | | `BRPHM_RUL_standard` | 1,151 | 997,506,805 bytes | Simulation development / 仿真开发 | | `BRPHM_RUL_complete` | 67,539 | 107,243,597,734 bytes | Full lineage research / 完整谱系研究 | All tiers are self-contained. The complete tier is the authoritative archival release; mini is intentionally not a statistical benchmark. / 每档均可独立使用；完整档是权威归档版本，mini 有意不作为统计基准。 ## What Is Included / 收录内容 The complete tier preserves the eligible original file formats under `raw/` and the current normalized research layer under `interim/`. Raw files are not silently transformed or discarded. Every normalized partition is a unit-level Parquet collection with its original `meta.json` retained. / 完整档在 `raw/` 保留合格的原始文件格式，在 `interim/` 保留当前规范化研究层；原始文件不会被静默转换或丢弃，每个规范化分区均保留单元级 Parquet 与原始 `meta.json`。 ### Source-To-Interim Lineage / 原始到规范化谱系 `standardized_to_interim` means a source has a current normalized partition. `raw_only` means the raw source is preserved but is not claimed to have been standardized. `simulation_canonical` identifies canonical internally generated simulation sources. / `standardized_to_interim` 表示已有当前规范化分区；`raw_only` 表示保留原始来源但绝不声称已规范化；`simulation_canonical` 表示规范的自建仿真来源。 | Source family / 来源族 | Raw path / 原始路径 | Interim / 规范化分区 | Status / 状态 | | --- | --- | --- | --- | | `A1_nasa_pcoe` | `raw/public/A1_nasa_pcoe` | `A1_nasa_pcoe` | `standardized_to_interim` | | `A2_nasa_random` | `raw/public/A2_nasa_random` | `A2_nasa_random` | `standardized_to_interim` | | `A3_nasa_smallsat` | `raw/public/A3_nasa_smallsat` | `A3_nasa_smallsat` | `standardized_to_interim` | | `A4_ucf_rr` | `raw/public/A4_ucf_rr` | `A4_ucf_rr` | `standardized_to_interim` | | `B1_matr` | `raw/public/B1_matr` | `B1_matr` | `standardized_to_interim` | | `B2_hust` | `raw/public/B2_hust` | `B2_hust` | `standardized_to_interim` | | `B3` | `raw/public/B3` | `-` | `raw_only` | | `B4` | `raw/public/B4` | `-` | `raw_only` | | `B5` | `raw/public/B5` | `-` | `raw_only` | | `C1` | `raw/public/C1` | `-` | `raw_only` | | `C4` | `raw/public/C4` | `-` | `raw_only` | | `D1_ims` | `raw/public/D1_ims` | `D1_ims` | `standardized_to_interim` | | `D2_femto` | `raw/public/D2_femto` | `D2_femto` | `standardized_to_interim` | | `D3_xjtu_sy` | `raw/public/D3_xjtu_sy` | `D3_xjtu_sy` | `standardized_to_interim` | | `D5_paderborn` | `raw/public/D5_paderborn` | `D5_paderborn` | `standardized_to_interim` | | `E2` | `raw/public/E2` | `-` | `raw_only` | | `F1_cmapss` | `raw/public/F1_cmapss` | `F1_cmapss` | `standardized_to_interim` | | `F2` | `raw/public/F2` | `-` | `raw_only` | | `SIM_bat` | `raw/sim/bat` | `SIM_bat` | `simulation_canonical` | | `SIM_dev` | `raw/sim/dev` | `SIM_dev` | `simulation_canonical` | | `SIM_rwa` | `raw/sim/rwa` | `SIM_rwa` | `simulation_canonical` | ## Simulation Construction / 仿真构建 `SIM_bat` and `SIM_rwa` are controlled, internally generated scenario trajectories. The release keeps the original MATLAB outputs, normalized partition metadata, seed/configuration fingerprints, source checksums, and the observed parser snapshot together. They are controlled simulation records, not a substitute for flight qualification data. / `SIM_bat` 和 `SIM_rwa` 是可控的自建仿真场景轨迹。发布包同时保留原始 MATLAB 输出、规范化分区元数据、随机种子/配置指纹、源文件校验和解析器快照。它们是可控仿真记录，不替代飞行鉴定数据。 ### Operating Scenarios / 运行工况 - Each unit is bound to a design-of-experiments record containing the product line, LEO altitude (`LEO500`, `LEO550`, or `LEO700`), solar beta angle, heat level, load level, series identifier, and random seed. Repeated orbital schedules supply environment state; the reaction-wheel records expose eclipse state and beta angle as environment fields. / 每个单元绑定一条试验设计记录，包含部件类型、近地轨道高度（`LEO500`、`LEO550` 或 `LEO700`）、太阳 β 角、热等级、负载等级、序列号和随机种子。重复轨道日程提供环境状态；反作用轮记录中将日蚀状态和 β 角作为环境字段。 - Battery trajectories are summarized once per orbital settlement cycle. Public telemetry includes capacity, mean temperature, charge duration, and mean C-rate. / 电池轨迹按每个轨道结算周期汇总；公开遥测包括容量、平均温度、充电时长和平均倍率。 - Reaction-wheel trajectories use the recorded native 0.1 Hz (10 s) grid. Public telemetry includes speed, motor current, commanded torque, and bearing temperature. / 反作用轮轨迹使用记录的 0.1 Hz（10 s）原生栅格；公开遥测包括轮速、电机电流、指令力矩和轴承温度。 ### Fault And Long-Term Degradation / 故障与长期衰减 - Battery aging carries simulator state for state of health (SOH), loss of lithium inventory (LLI), and loss of active material (LAM). The released failure label begins at the first cycle with `SOH <= 0.80`; the simulator stop-tail setting is explicitly separate from that label criterion. / 电池老化保留健康度（SOH）、锂损失（LLI）和活性材料损失（LAM）的仿真状态。已发布的故障标签从首个 `SOH <= 0.80` 的循环开始；仿真停止尾部设置与该标签判据明确分离。 - Reaction-wheel degradation accumulates wear, lubricant state, and friction state. Failure is defined by the first linear crossing of wear volume `V >= 2 mm^3`, or a speed-tracking error above 150 rpm sustained for 30 s in the 1 Hz supervisory post-processing. / 反作用轮衰减累积磨损、润滑状态和摩擦状态。故障定义为磨损体积 `V >= 2 mm^3` 的首次线性穿越，或在 1 Hz 监督后处理中速度跟踪误差超过 150 rpm 并持续 30 s。 - Long-horizon health states are retained only as `privileged` reference fields; observable channels and `label.*` targets remain separately declared in each partition `meta.json`. RUL is recorded without an artificial cap, while HI and failure labels use the stated partition-specific definitions. / 长时域健康状态仅以 `privileged` 参考字段保留；可观测通道与 `label.*` 目标均在各分区 `meta.json` 中单独声明。RUL 不人为封顶，HI 与故障标签遵循已声明的分区特定定义。 ### Traceability And Reproduction / 追溯与再生成 For each simulation unit, `meta.json` records the seed, scenario overrides, configuration YAML SHA-256, simulator and output-schema versions, environment reference, MATLAB version, creation time, and source-MAT SHA-256. The associated conversion logic is preserved under `processing/source_snapshot/src/datasets/sim_loader.py`, and the simulation dictionary is at `processing/source_snapshot/configs/sim/data_dictionary.json`. These records make the generated scenario auditable; they do not assert an independent validation of the underlying physical model. / 每个仿真单元的 `meta.json` 记录随机种子、场景覆盖参数、配置 YAML 的 SHA-256、仿真器与输出模式版本、环境引用、MATLAB 版本、生成时间和源 MAT 的 SHA-256。对应转换逻辑保存在 `processing/source_snapshot/src/datasets/sim_loader.py`，仿真字典位于 `processing/source_snapshot/configs/sim/data_dictionary.json`。这些记录使生成场景可审计，但不声称对底层物理模型的独立验证。 ## Processing Contract / 处理契约 1. **Preserve / 保留:** eligible raw records are copied byte-for-byte into the corresponding `raw/` tier path. 2. **Normalize / 规范化:** the current `interim/` layer is retained as unit-level Parquet; parser snapshots and the simulation data dictionary are under `processing/source_snapshot/`. 3. **Describe / 描述:** `PROVENANCE.json` maps each source family to its normalized status, observed citation, URL, and parser module when available. 4. **Verify / 验证:** every released regular file has a SHA-256 entry and is checked against its source during construction and independent verification. Raw-only sources are valuable archival evidence, but they are not interchangeable with normalized task-ready Parquet. / 仅原始来源具有归档价值，但不能与任务就绪的规范化 Parquet 混为一谈。 ## Data Quality And Integrity / 数据质量与完整性 | Observed statistic / 实测统计 | Value / 数值 | | --- | ---: | | Raw files in complete tier / 完整档原始文件 | 65,032 | | Raw bytes in complete tier / 完整档原始字节 | 107,078,917,085 | | Parquet files / Parquet 文件 | 2,493 | | Parquet rows / Parquet 行数 | 2,890,925 | | Time indices checked / 已检查时间索引 | 2,493 | | Non-monotonic time indices / 非单调时间索引 | 0 | | Parquet parse errors / Parquet 解析错误 | 0 | Machine evidence is in root `MANIFEST.json`, per-tier `MANIFEST.json`, `PROVENANCE.json`, `QUALITY_SUMMARY.json`, and `processing/MANIFEST.json`. These are machine records, not additional reading requirements. / 机器证据位于各 JSON 清单中，不增加人读文档负担。 ## Schema And Visibility / 模式与字段可见性 `t` is the time index. `label.*` fields are supervision labels. Fields declared `telemetry` in a partition `meta.json` are observation inputs. Fields declared `privileged` are research-only latent or reference signals and must not be represented as deployment-time telemetry. `/` 分区 `meta.json` 是字段、单位和可见性的唯一权威定义。 ## Reproducibility / 可复现性 Use `DEVELOPER_GUIDE.md` for a loading example and manifest verification. The release does not bundle processed training tensors, held-out evaluation data, trained models, or benchmark claims; therefore it is a data release, not a claim about model accuracy or cross-domain generalization. / 本发布是数据发布，不等同于模型精度或泛化结论。 ## Intended Use And Limits / 预期用途与边界 - Appropriate: dataset inspection, parser validation, feature research, source-aware RUL studies, and reproducible preprocessing work. / 适用于数据核验、解析器验证、特征研究、来源感知 RUL 研究和可复现预处理。 - Not appropriate: recovering withheld evaluation material, treating raw-only sources as normalized task labels, or inferring production telemetry from privileged fields. / 不得用于恢复封存评测数据、将仅原始来源当作规范化任务标签，或把特权字段当作部署遥测。 ## Citation / 引用 Please cite this collection by name and cite each external source according to the observed `license_cite` field in `PROVENANCE.json` and the original source artifacts. / 请引用本集合，并按 `PROVENANCE.json` 与原始来源工件中的引用要求分别引用外部来源。 ## Publication Tiers / 发布档位 This repository is published monotonically. The current public data tier is **BRPHM_RUL_complete**; all three declared tiers are published, and no unpublished future tier is reserved. 本仓库按递进阶段发布。当前公开数据档位为 **BRPHM_RUL_complete**；三个声明档位的数据均已上传，不预设未发布的后续档位。 ## Immutable Docker Compatibility / 固定 Docker 兼容性 The tier data are intended to be acquired outside the immutable application image and mounted read-only for processing. With the fixed `brphm-preview` image digest `sha256:43ec2100ef903d5dca88d5e0e928c4c0bbba1297cf1a241203f079b2357380a4`, run the application with an empty `PYTHONPATH` when using `reconstruct`; this preserves the image files and avoids a namespace collision in the image's default environment. The standard-tier read-only smoke test covered manifest verification and one BAT reconstruction with `--verify`. Any future dataset change must repeat full manifest-file verification and this fixed-image gate before publication. 固定应用镜像之外获取数据集，并将档位以只读方式挂载后处理。使用固定 `brphm-preview` 镜像摘要 `sha256:43ec2100ef903d5dca88d5e0e928c4c0bbba1297cf1a241203f079b2357380a4` 时，执行 `reconstruct` 应设置空的 `PYTHONPATH`；这不会改动镜像文件，并可避免镜像默认环境中的命名空间冲突。standard 档只读冒烟测试已覆盖清单校验和一个带 `--verify` 的 BAT 重建。以后任何数据集修改在发布前都必须重新通过逐文件清单校验和该固定镜像门禁。