File size: 63,160 Bytes
9f8bd1b 1a3194d 9f8bd1b 1a3194d 9f8bd1b 1a3194d 9f8bd1b 1a3194d 9f8bd1b 1a3194d 9f8bd1b 1a3194d 9f8bd1b 1a3194d 9f8bd1b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 | import os
import io
import math
import zipfile
from pathlib import Path
from typing import Dict, Optional, Tuple
from urllib.parse import quote
import numpy as np
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
import streamlit as st
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.impute import SimpleImputer
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, confusion_matrix
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
APP_DIR = Path(__file__).resolve().parent
DEFAULT_DATA_DIR = APP_DIR / "data" / "processed"
OUTPUT_FIG_DIR = APP_DIR / "outputs" / "figures"
OUTPUT_TABLE_DIR = APP_DIR / "outputs" / "tables"
RAW_DATA_DIR = APP_DIR / "data" / "raw"
DEFAULT_RAW_TELEMETRY_PATH = RAW_DATA_DIR / "generator_telemetry_with_labels.csv"
# Hugging Face Spaces can store larger files through Xet/LFS. In some Docker builds,
# the local file may appear as a tiny pointer file instead of the real CSV.
# This fallback lets the app fetch the resolved file from the Space repo when needed.
HF_SPACE_ID = os.getenv(
"HF_SPACE_ID",
"AnubhaParashar/Predictive_Preventive_Maintenance_for_Generator_Reliability",
)
HF_RESOLVE_BASE = f"https://huggingface.co/spaces/{HF_SPACE_ID}/resolve/main"
EXTERNAL_RAW_TELEMETRY_PATH = DEFAULT_RAW_TELEMETRY_PATH
# Snowflake Streamlit does not support page_title/page_icon in st.set_page_config.
# Keep layout only and continue safely if the runtime ignores this call.
try:
st.set_page_config(layout="wide", initial_sidebar_state="expanded")
except Exception:
pass
# -----------------------------
# Utility helpers
# -----------------------------
def money(x):
try:
if pd.isna(x):
return "$0"
return f"${float(x):,.0f}"
except Exception:
return "$0"
def num(x, decimals=0):
try:
if pd.isna(x):
return "0"
return f"{float(x):,.{decimals}f}"
except Exception:
return "0"
def pct(x, decimals=1):
try:
if pd.isna(x):
return "0%"
return f"{float(x) * 100:.{decimals}f}%"
except Exception:
return "0%"
def _looks_like_xet_or_lfs_pointer(df: pd.DataFrame) -> bool:
"""Detect a Git LFS / Hugging Face Xet pointer that was read as a tiny CSV."""
if df is None or df.empty:
return False
if df.shape[0] > 5 or df.shape[1] > 2:
return False
text = " ".join([str(x) for x in list(df.columns) + df.astype(str).values.flatten().tolist()])
text = text.lower()
return (
"git-lfs.github.com/spec" in text
or "version https://git-lfs" in text
or "xet" in text and "sha256" in text
or "oid sha256" in text
)
def _hf_resolve_url_for_path(path: Path) -> str:
"""Build a Hugging Face /resolve/main URL for a local repo-relative path."""
try:
rel = path.resolve().relative_to(APP_DIR.resolve()).as_posix()
except Exception:
rel = path.as_posix().lstrip("/")
return f"{HF_RESOLVE_BASE}/{quote(rel)}"
def read_csv_safely(path_or_buffer) -> pd.DataFrame:
"""Read CSV and recover automatically if Hugging Face gives a Xet/LFS pointer file."""
df = pd.read_csv(path_or_buffer)
# If a local repo file is only a pointer, fetch the actual resolved file from Hugging Face.
if isinstance(path_or_buffer, (str, Path)):
path = Path(path_or_buffer)
if _looks_like_xet_or_lfs_pointer(df):
resolved_url = _hf_resolve_url_for_path(path)
return pd.read_csv(resolved_url)
return df
def parse_date_cols(df: pd.DataFrame, cols) -> pd.DataFrame:
out = df.copy()
for c in cols:
if c in out.columns:
out[c] = pd.to_datetime(out[c], errors="coerce")
return out
@st.cache_data(show_spinner=False)
def load_default_tables() -> Dict[str, pd.DataFrame]:
tables = {}
files = {
"asset_master": "asset_master.csv",
"pm_events": "pm_events.csv",
"failure_events": "failure_events.csv",
"business_impact": "business_impact.csv",
"pm_failure_linked": "pm_failure_linked.csv",
"telemetry_weekly": "telemetry_weekly.csv",
}
for key, filename in files.items():
path = DEFAULT_DATA_DIR / filename
if path.exists():
tables[key] = read_csv_safely(path)
else:
tables[key] = pd.DataFrame()
tables["asset_master"] = parse_date_cols(tables["asset_master"], ["install_date"])
tables["pm_events"] = parse_date_cols(tables["pm_events"], ["scheduled_date", "completed_date", "pm_date"])
tables["failure_events"] = parse_date_cols(
tables["failure_events"], ["failure_date", "ticket_open_date", "ticket_close_date"]
)
tables["business_impact"] = parse_date_cols(tables["business_impact"], ["event_date"])
tables["pm_failure_linked"] = parse_date_cols(tables["pm_failure_linked"], ["pm_date", "next_failure_date"])
tables["telemetry_weekly"] = parse_date_cols(tables["telemetry_weekly"], ["timestamp"])
return tables
@st.cache_data(show_spinner=False)
def load_default_raw_telemetry() -> Tuple[pd.DataFrame, str]:
"""Load the packaged raw telemetry file used by the demo app."""
if DEFAULT_RAW_TELEMETRY_PATH.exists():
df = read_csv_safely(DEFAULT_RAW_TELEMETRY_PATH)
return parse_date_cols(df, ["timestamp"]), "Packaged demo file: data/raw/generator_telemetry_with_labels.csv"
return pd.DataFrame(), "No packaged raw telemetry file found"
def dataframe_profile(df: pd.DataFrame, source_path: str) -> pd.DataFrame:
if df.empty:
return pd.DataFrame({"Metric": ["Source path", "Rows", "Columns"], "Value": [source_path, "0", "0"]})
values = {
"Source path": source_path,
"Rows": f"{len(df):,}",
"Columns": f"{len(df.columns):,}",
"Date range": "Not available",
"Duplicate rows": f"{int(df.duplicated().sum()):,}",
"Missing values": f"{int(df.isna().sum().sum()):,}",
}
if "timestamp" in df.columns:
ts = pd.to_datetime(df["timestamp"], errors="coerce")
if ts.notna().any():
values["Date range"] = f"{ts.min().date()} to {ts.max().date()}"
return pd.DataFrame({"Metric": list(values.keys()), "Value": list(values.values())})
def enrich_algorithm_output(raw_df: pd.DataFrame) -> pd.DataFrame:
"""Create a transparent demo algorithm output from raw telemetry columns."""
if raw_df.empty:
return pd.DataFrame()
out = raw_df.copy()
if "timestamp" in out.columns:
out["timestamp"] = pd.to_datetime(out["timestamp"], errors="coerce")
for c in ["anomaly_score", "days_since_last_pm", "failure_within_30d", "failure_within_14d"]:
if c in out.columns:
out[c] = pd.to_numeric(out[c], errors="coerce")
if "anomaly_score" not in out.columns:
numeric_cols = out.select_dtypes(include=[np.number]).columns.tolist()
if numeric_cols:
z = out[numeric_cols].fillna(out[numeric_cols].median(numeric_only=True))
out["anomaly_score"] = ((z - z.mean()) / z.std(ddof=0)).abs().mean(axis=1).rank(pct=True) * 100
else:
out["anomaly_score"] = 0
conditions = [
out["anomaly_score"] >= 80,
out["anomaly_score"] >= 60,
out["anomaly_score"] >= 40,
]
choices = ["Critical", "High", "Watch"]
out["algorithm_risk_band"] = np.select(conditions, choices, default="Low")
pm_age = pd.to_numeric(out.get("days_since_last_pm", pd.Series(0, index=out.index)), errors="coerce").fillna(0)
out["algorithm_failure_risk_score"] = np.clip((out["anomaly_score"].fillna(0) * 0.75) + np.minimum(pm_age, 180) / 180 * 25, 0, 100).round(2)
out["algorithm_predicted_failure_within_30d"] = (out["algorithm_failure_risk_score"] >= 70).astype(int)
out["recommended_action"] = np.select(
[out["algorithm_risk_band"].eq("Critical"), out["algorithm_risk_band"].eq("High"), out["algorithm_risk_band"].eq("Watch")],
["Immediate inspection / corrective work order", "Schedule PM within 7 days", "Monitor and plan PM",],
default="Normal monitoring",
)
sort_cols = [c for c in ["timestamp", "algorithm_failure_risk_score"] if c in out.columns]
if sort_cols:
out = out.sort_values(sort_cols, ascending=[False, False] if len(sort_cols) == 2 else False)
return out
def summarize_algorithm_by_asset(algo_df: pd.DataFrame) -> pd.DataFrame:
if algo_df.empty or "asset_id" not in algo_df.columns:
return pd.DataFrame()
df = algo_df.copy()
if "timestamp" in df.columns:
df = df.sort_values("timestamp")
latest = df.groupby("asset_id", as_index=False).tail(1).copy()
keep_cols = [c for c in ["asset_id", "timestamp", "model", "region", "criticality", "environment_type", "days_since_last_pm", "anomaly_score", "algorithm_failure_risk_score", "algorithm_risk_band", "algorithm_predicted_failure_within_30d", "recommended_action"] if c in latest.columns]
return latest[keep_cols].sort_values("algorithm_failure_risk_score", ascending=False)
# -----------------------------
# Machine Learning helpers
# -----------------------------
@st.cache_data(show_spinner=False)
def prepare_ml_frame(raw_df: pd.DataFrame, target_col: str = "failure_within_30d", max_rows: int = 90000):
"""Prepare telemetry data for supervised failure-risk modeling."""
if raw_df.empty or target_col not in raw_df.columns:
return pd.DataFrame(), [], []
use_cols = [
"asset_id", "age_years", "days_since_last_pm", "runtime_hours_week", "avg_load_pct", "oil_temp_c",
"coolant_temp_c", "battery_voltage", "vibration_mm_s", "fuel_rate_lph", "alarm_count", "anomaly_score",
"model", "region", "criticality", "environment_type", target_col, "timestamp"
]
use_cols = [c for c in use_cols if c in raw_df.columns]
df = raw_df[use_cols].copy().dropna(subset=[target_col])
if "timestamp" in df.columns:
df["timestamp"] = pd.to_datetime(df["timestamp"], errors="coerce")
df = df.sort_values("timestamp")
# Keep the app fast in Snowflake by training on a balanced sample.
if len(df) > max_rows:
pos = df[df[target_col] == 1]
neg = df[df[target_col] == 0]
n_pos = min(len(pos), max_rows // 3)
n_neg = max_rows - n_pos
df = pd.concat([
pos.sample(n=n_pos, random_state=42) if len(pos) > n_pos else pos,
neg.sample(n=n_neg, random_state=42) if len(neg) > n_neg else neg,
], axis=0).sample(frac=1, random_state=42)
feature_cols = [c for c in df.columns if c not in [target_col, "timestamp"]]
numeric_cols = [c for c in feature_cols if pd.api.types.is_numeric_dtype(df[c])]
categorical_cols = [c for c in feature_cols if c not in numeric_cols]
return df, numeric_cols, categorical_cols
@st.cache_resource(show_spinner=False)
def train_failure_models(df: pd.DataFrame, numeric_cols, categorical_cols, target_col: str = "failure_within_30d"):
"""Train baseline and advanced ML classifiers for near-term failure prediction."""
X = df[numeric_cols + categorical_cols]
y = df[target_col].astype(int)
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.25, random_state=42, stratify=y
)
numeric_transformer = Pipeline(steps=[
("imputer", SimpleImputer(strategy="median")),
("scaler", StandardScaler()),
])
categorical_transformer = Pipeline(steps=[
("imputer", SimpleImputer(strategy="most_frequent")),
("onehot", OneHotEncoder(handle_unknown="ignore")),
])
preprocessor = ColumnTransformer(
transformers=[
("num", numeric_transformer, numeric_cols),
("cat", categorical_transformer, categorical_cols),
]
)
models = {
"Logistic Regression baseline": LogisticRegression(max_iter=500, class_weight="balanced"),
"Random Forest advanced model": RandomForestClassifier(
n_estimators=160,
max_depth=12,
min_samples_leaf=5,
random_state=42,
n_jobs=-1,
class_weight="balanced_subsample",
),
}
results = {}
for name, model in models.items():
pipe = Pipeline(steps=[("preprocessor", preprocessor), ("model", model)])
pipe.fit(X_train, y_train)
proba = pipe.predict_proba(X_test)[:, 1]
pred = (proba >= 0.5).astype(int)
results[name] = {
"pipeline": pipe,
"accuracy": accuracy_score(y_test, pred),
"precision": precision_score(y_test, pred, zero_division=0),
"recall": recall_score(y_test, pred, zero_division=0),
"f1": f1_score(y_test, pred, zero_division=0),
"roc_auc": roc_auc_score(y_test, proba),
"confusion_matrix": confusion_matrix(y_test, pred),
}
return results
def get_ml_feature_importance(pipeline, model_name: str, top_n: int = 15) -> pd.DataFrame:
preprocessor = pipeline.named_steps["preprocessor"]
feature_names = preprocessor.get_feature_names_out()
model = pipeline.named_steps["model"]
if "Logistic" in model_name:
values = np.abs(model.coef_[0])
else:
values = model.feature_importances_
out = pd.DataFrame({"feature": feature_names, "importance": values})
out["feature"] = out["feature"].str.replace("num__", "", regex=False).str.replace("cat__", "", regex=False)
return out.sort_values("importance", ascending=False).head(top_n)
def score_latest_asset_risk(telemetry_df: pd.DataFrame, pipeline) -> pd.DataFrame:
latest = telemetry_df.copy()
if latest.empty or "asset_id" not in latest.columns:
return pd.DataFrame()
if "timestamp" in latest.columns:
latest["timestamp"] = pd.to_datetime(latest["timestamp"], errors="coerce")
latest = latest.sort_values("timestamp").groupby("asset_id", as_index=False).tail(1)
feature_cols = [
"asset_id", "age_years", "days_since_last_pm", "runtime_hours_week", "avg_load_pct", "oil_temp_c",
"coolant_temp_c", "battery_voltage", "vibration_mm_s", "fuel_rate_lph", "alarm_count", "anomaly_score",
"model", "region", "criticality", "environment_type"
]
feature_cols = [c for c in feature_cols if c in latest.columns]
latest = latest.copy()
latest["ml_predicted_failure_risk"] = pipeline.predict_proba(latest[feature_cols])[:, 1]
latest["ml_risk_band"] = pd.cut(
latest["ml_predicted_failure_risk"],
bins=[-0.01, 0.30, 0.60, 0.80, 1.0],
labels=["Low", "Medium", "High", "Critical"],
)
latest["ml_recommended_action"] = np.select(
[
latest["ml_predicted_failure_risk"] >= 0.80,
latest["ml_predicted_failure_risk"] >= 0.60,
latest["ml_predicted_failure_risk"] >= 0.30,
],
[
"Immediate inspection / PM within 3 days",
"Schedule PM within 7 days",
"Watch list / inspect soon",
],
default="Normal monitoring",
)
keep = [c for c in [
"asset_id", "timestamp", "model", "region", "criticality", "days_since_last_pm",
"anomaly_score", "battery_voltage", "vibration_mm_s", "alarm_count",
"ml_predicted_failure_risk", "ml_risk_band", "ml_recommended_action"
] if c in latest.columns]
return latest[keep].sort_values("ml_predicted_failure_risk", ascending=False)
def build_prescriptive_pm_plan(ml_risk_df: pd.DataFrame, linked_df: pd.DataFrame, business_df: pd.DataFrame) -> pd.DataFrame:
if ml_risk_df.empty:
return pd.DataFrame()
plan = ml_risk_df.copy()
if not linked_df.empty and {"asset_id", "days_to_next_failure"}.issubset(linked_df.columns):
hist = linked_df.copy()
hist["days_to_next_failure"] = pd.to_numeric(hist["days_to_next_failure"], errors="coerce")
hist = hist.groupby("asset_id", as_index=False)["days_to_next_failure"].median()
hist = hist.rename(columns={"days_to_next_failure": "historical_median_days_to_failure"})
plan = plan.merge(hist, on="asset_id", how="left")
avg_failure_cost = 0.0
if not business_df.empty:
if "repair_cost" in business_df.columns:
avg_failure_cost += pd.to_numeric(business_df["repair_cost"], errors="coerce").fillna(0).mean()
if "truck_roll_cost" in business_df.columns:
avg_failure_cost += pd.to_numeric(business_df["truck_roll_cost"], errors="coerce").fillna(0).mean()
plan["estimated_failure_cost_exposure"] = avg_failure_cost
plan["prescriptive_pm_action"] = np.select(
[
plan["ml_predicted_failure_risk"] >= 0.80,
(plan["ml_predicted_failure_risk"] >= 0.60) | (pd.to_numeric(plan.get("days_since_last_pm", 0), errors="coerce").fillna(0) > 120),
plan["ml_predicted_failure_risk"] >= 0.30,
],
[
"Create urgent work order and inspect immediately",
"Plan PM in next 7 days and review alarms / vibration",
"Increase monitoring and inspect in next PM window",
],
default="Maintain current PM cycle",
)
plan["priority_rank"] = plan["ml_predicted_failure_risk"].rank(method="dense", ascending=False).astype(int)
return plan.sort_values("priority_rank")
def identify_uploaded_zip(uploaded_zip) -> Dict[str, pd.DataFrame]:
"""Accepts a zip containing CSVs with expected names."""
tables = {}
if uploaded_zip is None:
return tables
expected = {
"asset_master": "asset_master.csv",
"pm_events": "pm_events.csv",
"failure_events": "failure_events.csv",
"business_impact": "business_impact.csv",
"pm_failure_linked": "pm_failure_linked.csv",
"telemetry_weekly": "telemetry_weekly.csv",
}
with zipfile.ZipFile(uploaded_zip) as z:
names = z.namelist()
for key, filename in expected.items():
match = [n for n in names if n.endswith(filename)]
if match:
with z.open(match[0]) as f:
tables[key] = pd.read_csv(f)
return tables
def build_pm_failure_linked(pm_df: pd.DataFrame, failure_df: pd.DataFrame) -> pd.DataFrame:
required_pm = {"pm_event_id", "asset_id", "pm_date"}
required_fail = {"failure_event_id", "asset_id", "failure_date"}
if pm_df.empty or failure_df.empty:
return pd.DataFrame()
if not required_pm.issubset(pm_df.columns) or not required_fail.issubset(failure_df.columns):
return pd.DataFrame()
pm = parse_date_cols(pm_df, ["pm_date"]).copy()
failures = parse_date_cols(failure_df, ["failure_date"]).copy()
failure_groups = {}
for asset_id, g in failures.dropna(subset=["failure_date"]).sort_values("failure_date").groupby("asset_id"):
failure_groups[asset_id] = g.reset_index(drop=True)
rows = []
for _, row in pm.dropna(subset=["pm_date"]).iterrows():
asset_id = row["asset_id"]
pm_date = row["pm_date"]
next_fail = None
if asset_id in failure_groups:
g = failure_groups[asset_id]
pos = np.searchsorted(g["failure_date"].values.astype("datetime64[ns]"), np.datetime64(pm_date), side="right")
if pos < len(g):
next_fail = g.iloc[int(pos)]
if next_fail is not None:
days = (next_fail["failure_date"] - pm_date).days
rows.append({
"pm_event_id": row.get("pm_event_id"),
"asset_id": asset_id,
"pm_date": pm_date,
"next_failure_event_id": next_fail.get("failure_event_id"),
"next_failure_date": next_fail.get("failure_date"),
"days_to_next_failure": days,
"failure_found_flag": 1,
"ontime_flag": row.get("ontime_flag", np.nan),
"delay_days": row.get("delay_days", np.nan),
"pm_total_cost": row.get("total_cost", np.nan),
"failure_total_cost": next_fail.get("total_cost", np.nan),
"failure_category": next_fail.get("failure_category", "Unknown"),
})
else:
rows.append({
"pm_event_id": row.get("pm_event_id"),
"asset_id": asset_id,
"pm_date": pm_date,
"next_failure_event_id": None,
"next_failure_date": pd.NaT,
"days_to_next_failure": np.nan,
"failure_found_flag": 0,
"ontime_flag": row.get("ontime_flag", np.nan),
"delay_days": row.get("delay_days", np.nan),
"pm_total_cost": row.get("total_cost", np.nan),
"failure_total_cost": np.nan,
"failure_category": None,
})
return pd.DataFrame(rows)
def apply_asset_filters(tables: Dict[str, pd.DataFrame], regions, models, criticalities) -> Dict[str, pd.DataFrame]:
assets = tables.get("asset_master", pd.DataFrame()).copy()
if assets.empty:
return tables
mask = pd.Series(True, index=assets.index)
if regions and "region" in assets.columns:
mask &= assets["region"].isin(regions)
if models and "model" in assets.columns:
mask &= assets["model"].isin(models)
if criticalities and "criticality" in assets.columns:
mask &= assets["criticality"].isin(criticalities)
keep_assets = set(assets.loc[mask, "asset_id"].astype(str))
out = {}
for key, df in tables.items():
if isinstance(df, pd.DataFrame) and not df.empty and "asset_id" in df.columns:
out[key] = df[df["asset_id"].astype(str).isin(keep_assets)].copy()
else:
out[key] = df.copy() if isinstance(df, pd.DataFrame) else df
return out
def survival_curve(linked: pd.DataFrame) -> pd.DataFrame:
if linked.empty or "days_to_next_failure" not in linked.columns:
return pd.DataFrame(columns=["days", "failure_free_probability"])
d = linked.loc[linked["days_to_next_failure"].notna(), "days_to_next_failure"].astype(float)
if d.empty:
return pd.DataFrame(columns=["days", "failure_free_probability"])
days_grid = np.arange(0, max(30, int(d.max()) + 30), 30)
n = len(d)
vals = []
for day in days_grid:
vals.append({
"days": int(day),
"failure_free_probability": float((d > day).sum() / n)
})
return pd.DataFrame(vals)
def make_download_zip(tables: Dict[str, pd.DataFrame]) -> bytes:
mem = io.BytesIO()
with zipfile.ZipFile(mem, "w", compression=zipfile.ZIP_DEFLATED) as z:
for name, df in tables.items():
if isinstance(df, pd.DataFrame) and not df.empty:
z.writestr(f"{name}.csv", df.to_csv(index=False))
mem.seek(0)
return mem.read()
def metric_card(label, value, help_text=None):
st.metric(label, value, help=help_text)
# -----------------------------
# Sidebar: data loading
# -----------------------------
st.sidebar.title("⚙️ Dashboard Controls")
st.sidebar.markdown(
"""
**Data mode**
- Default mode loads the included generator PM demo dataset.
- Upload mode lets you override one or more CSVs.
"""
)
data_mode = st.sidebar.radio(
"Choose data source",
["Use included default dataset", "Upload my own CSVs / ZIP"],
index=0,
)
tables = load_default_tables()
raw_telemetry_df, raw_telemetry_source_path = load_default_raw_telemetry()
if data_mode == "Upload my own CSVs / ZIP":
st.sidebar.info("Upload a ZIP with expected CSV names, or upload individual CSVs below. Missing files fall back to default data.")
uploaded_zip = st.sidebar.file_uploader(
"Optional: upload full data ZIP",
type=["zip"],
help="ZIP may contain asset_master.csv, pm_events.csv, failure_events.csv, telemetry_weekly.csv, business_impact.csv, pm_failure_linked.csv",
)
try:
zip_tables = identify_uploaded_zip(uploaded_zip)
for k, v in zip_tables.items():
tables[k] = v
except Exception as e:
st.sidebar.error(f"Could not read ZIP: {e}")
upload_specs = {
"asset_master": "asset_master.csv",
"pm_events": "pm_events.csv",
"failure_events": "failure_events.csv",
"business_impact": "business_impact.csv",
"pm_failure_linked": "pm_failure_linked.csv",
"telemetry_weekly": "telemetry_weekly.csv",
}
with st.sidebar.expander("Upload individual CSVs"):
for key, label in upload_specs.items():
f = st.file_uploader(label, type=["csv"], key=f"upload_{key}")
if f is not None:
try:
tables[key] = pd.read_csv(f)
except Exception as e:
st.error(f"Could not read {label}: {e}")
with st.sidebar.expander("Upload raw telemetry CSV"):
raw_file = st.file_uploader(
"generator_telemetry_with_labels.csv",
type=["csv"],
key="upload_raw_telemetry",
help="Optional raw telemetry file before algorithm enrichment. If omitted, dashboard uses the WSL path when found, otherwise the included package file.",
)
if raw_file is not None:
try:
raw_telemetry_df = parse_date_cols(pd.read_csv(raw_file), ["timestamp"])
raw_telemetry_source_path = "Uploaded file: generator_telemetry_with_labels.csv"
except Exception as e:
st.error(f"Could not read raw telemetry CSV: {e}")
# Parse dates after custom upload.
tables["asset_master"] = parse_date_cols(tables.get("asset_master", pd.DataFrame()), ["install_date"])
tables["pm_events"] = parse_date_cols(tables.get("pm_events", pd.DataFrame()), ["scheduled_date", "completed_date", "pm_date"])
tables["failure_events"] = parse_date_cols(tables.get("failure_events", pd.DataFrame()), ["failure_date", "ticket_open_date", "ticket_close_date"])
tables["business_impact"] = parse_date_cols(tables.get("business_impact", pd.DataFrame()), ["event_date"])
tables["pm_failure_linked"] = parse_date_cols(tables.get("pm_failure_linked", pd.DataFrame()), ["pm_date", "next_failure_date"])
tables["telemetry_weekly"] = parse_date_cols(tables.get("telemetry_weekly", pd.DataFrame()), ["timestamp"])
raw_telemetry_df = parse_date_cols(raw_telemetry_df, ["timestamp"])
algorithm_output_df = enrich_algorithm_output(raw_telemetry_df)
algorithm_asset_summary_df = summarize_algorithm_by_asset(algorithm_output_df)
# If the user uploaded PM/failure but not linked, rebuild it.
if tables.get("pm_failure_linked", pd.DataFrame()).empty and not tables.get("pm_events", pd.DataFrame()).empty and not tables.get("failure_events", pd.DataFrame()).empty:
tables["pm_failure_linked"] = build_pm_failure_linked(tables["pm_events"], tables["failure_events"])
# Also allow forced rebuild.
if st.sidebar.button("Rebuild PM → Failure Links"):
tables["pm_failure_linked"] = build_pm_failure_linked(tables["pm_events"], tables["failure_events"])
st.sidebar.success("Rebuilt pm_failure_linked from PM and failure tables.")
assets = tables.get("asset_master", pd.DataFrame())
regions = []
models = []
criticalities = []
if not assets.empty:
if "region" in assets.columns:
all_regions = sorted([x for x in assets["region"].dropna().unique().tolist()])
regions = st.sidebar.multiselect("Filter region", all_regions, default=all_regions)
if "model" in assets.columns:
all_models = sorted([x for x in assets["model"].dropna().unique().tolist()])
models = st.sidebar.multiselect("Filter model", all_models, default=all_models)
if "criticality" in assets.columns:
all_criticality = sorted([x for x in assets["criticality"].dropna().unique().tolist()])
criticalities = st.sidebar.multiselect("Filter criticality", all_criticality, default=all_criticality)
sample_limit = st.sidebar.slider("Telemetry chart sample size", 1_000, 50_000, 10_000, step=1_000)
filtered = apply_asset_filters(tables, regions, models, criticalities)
asset_df = filtered.get("asset_master", pd.DataFrame())
pm_df = filtered.get("pm_events", pd.DataFrame())
failure_df = filtered.get("failure_events", pd.DataFrame())
business_df = filtered.get("business_impact", pd.DataFrame())
linked_df = filtered.get("pm_failure_linked", pd.DataFrame())
telemetry_df = filtered.get("telemetry_weekly", pd.DataFrame())
# -----------------------------
# Header
# -----------------------------
st.title("⚙️ Generator Preventive Maintenance Reliability Dashboard")
st.caption("Upload your own maintenance data or use the included demo dataset to analyze PM effectiveness, failure risk, and business impact.")
with st.expander("Expected CSV inputs", expanded=False):
st.markdown(
"""
**Recommended CSVs**
- `asset_master.csv`: one row per generator/asset
- `pm_events.csv`: preventive maintenance events
- `failure_events.csv`: unplanned repair/failure tickets
- `pm_failure_linked.csv`: optional; dashboard can rebuild it
- `telemetry_weekly.csv`: optional sensor/risk history
- `generator_telemetry_with_labels.csv`: optional ML training table with `failure_within_14d` and `failure_within_30d`
- `business_impact.csv`: optional downtime, repair cost, customer impact
**Minimum required for PM-to-failure analysis**
- `pm_events.csv` with `pm_event_id`, `asset_id`, `pm_date`
- `failure_events.csv` with `failure_event_id`, `asset_id`, `failure_date`
"""
)
# -----------------------------
# Tabs
# -----------------------------
tabs = st.tabs([
"Problem & Solution",
"Dataset & Raw → Algorithm",
"Executive KPIs",
"PM Effectiveness",
"Failures & Cost",
"Telemetry Risk",
"Predictive ML & PM Strategy",
"Saved Outputs",
"Data Explorer & Export",
])
# -----------------------------
# Problem & Solution tab
# -----------------------------
with tabs[0]:
st.header("Problem Statement")
st.markdown(
"""
Telecom and field-service teams need to know whether **preventive maintenance is actually preventing generator failures**.
The key business question is:
> After a PM is completed, how long does a generator typically run before the next failure or repair ticket?
Without this, teams cannot confidently answer:
- Which PMs are effective?
- Which assets are becoming risky?
- How much downtime and cost can be avoided?
- Which regions, models, or environments need more attention?
"""
)
st.header("Solution Provided")
st.markdown(
"""
This dashboard converts maintenance records into a reliability workflow:
1. **Load data**
Use the included generator demo dataset or upload your own CSVs.
2. **Link PM to next failure**
For every PM event, the dashboard finds the first later failure ticket for the same asset.
3. **Measure PM effectiveness**
It calculates `days_to_next_failure`, failure-free rate, and on-time vs delayed PM comparisons.
4. **Analyze failure cost and downtime**
It summarizes repair cost, truck-roll cost, downtime, SLA impact, and customer impact.
5. **Train ML failure-risk models**
Logistic Regression and Random Forest predict whether a generator may fail within 14 or 30 days.
6. **Prescribe PM actions**
ML risk scores are converted into priority bands and recommended maintenance actions.
7. **Export outputs**
KPI tables and processed datasets can be downloaded for Power BI, Excel, Streamlit sharing, or a client deck.
"""
)
st.subheader("Architecture")
st.code(
"""
PM Events + Failure Tickets + Asset Master
│
▼
PM → Next Failure Linker
│
├── PM effectiveness KPIs
├── Failure-free survival curve
├── On-time vs delayed PM comparison
├── Failure cost and downtime view
└── Telemetry risk overlay
""",
language="text",
)
st.subheader("Dataset currently loaded")
c1, c2, c3, c4, c5 = st.columns(5)
c1.metric("Assets", f"{len(asset_df):,}")
c2.metric("PM events", f"{len(pm_df):,}")
c3.metric("Failure tickets", f"{len(failure_df):,}")
c4.metric("Telemetry rows", f"{len(telemetry_df):,}")
c5.metric("Linked PM rows", f"{len(linked_df):,}")
# -----------------------------
# -----------------------------
# Dataset and Raw → Algorithm tab
# -----------------------------
with tabs[1]:
st.header("Dataset Used: Raw → Algorithm Output")
st.markdown(
"""
This page explains exactly what data is being used, where it is expected in WSL, what the raw file looks like before the algorithm, and what the algorithm produces after enrichment.
**Packaged raw dataset used by the app:**
"""
)
st.code("data/raw/generator_telemetry_with_labels.csv", language="text")
st.info(
f"Currently loaded raw telemetry source: {raw_telemetry_source_path}\n\n"
"The dashboard uses the packaged demo raw file. You can replace it by uploading your own raw telemetry CSV from the sidebar."
)
st.subheader("1) Raw dataset profile before algorithm")
st.dataframe(dataframe_profile(raw_telemetry_df, raw_telemetry_source_path), use_container_width=True)
if raw_telemetry_df.empty:
st.warning("Raw telemetry dataset is not available. Place generator_telemetry_with_labels.csv at the WSL path above or upload it from the sidebar.")
else:
c1, c2 = st.columns([2, 1])
with c1:
st.markdown("**Raw telemetry preview**")
st.dataframe(raw_telemetry_df.head(1000), use_container_width=True)
with c2:
missing = raw_telemetry_df.isna().sum().sort_values(ascending=False).head(12).reset_index()
missing.columns = ["column", "missing_count"]
fig = px.bar(missing, x="missing_count", y="column", orientation="h", title="Raw missing values by column")
st.plotly_chart(fig, use_container_width=True)
st.markdown("**Raw signal visualizations before algorithm**")
c3, c4 = st.columns(2)
with c3:
if "anomaly_score" in raw_telemetry_df.columns:
fig = px.histogram(raw_telemetry_df, x="anomaly_score", nbins=50, title="Raw anomaly score distribution")
st.plotly_chart(fig, use_container_width=True)
elif "runtime_hours_week" in raw_telemetry_df.columns:
fig = px.histogram(raw_telemetry_df, x="runtime_hours_week", nbins=50, title="Raw weekly runtime distribution")
st.plotly_chart(fig, use_container_width=True)
with c4:
signal_cols = [c for c in ["oil_temp_c", "coolant_temp_c", "battery_voltage", "vibration_mm_s", "fuel_rate_lph", "runtime_hours_week"] if c in raw_telemetry_df.columns]
if signal_cols and "timestamp" in raw_telemetry_df.columns:
sample_asset = raw_telemetry_df["asset_id"].iloc[0] if "asset_id" in raw_telemetry_df.columns else None
signal = signal_cols[0]
plot_df = raw_telemetry_df[raw_telemetry_df["asset_id"].eq(sample_asset)].copy() if sample_asset and "asset_id" in raw_telemetry_df.columns else raw_telemetry_df.head(200)
fig = px.line(plot_df.sort_values("timestamp"), x="timestamp", y=signal, title=f"Raw signal trend example: {signal}")
st.plotly_chart(fig, use_container_width=True)
st.subheader("2) Algorithm applied")
st.markdown(
"""
The dashboard creates a transparent demo algorithm layer from the raw telemetry:
- Uses `anomaly_score` as the main health signal.
- Adds PM-age pressure using `days_since_last_pm`.
- Produces `algorithm_failure_risk_score` on a 0–100 scale.
- Converts risk score into `Low`, `Watch`, `High`, or `Critical` bands.
- Adds `algorithm_predicted_failure_within_30d` and a recommended maintenance action.
This is intentionally explainable for a business demo. When real client data is available, this layer can be replaced by a trained classifier, survival model, or remaining-useful-life model.
"""
)
st.code(
"""
algorithm_failure_risk_score = 0.75 * anomaly_score + 0.25 * normalized_days_since_last_pm
risk bands:
Low < 40
Watch 40–59
High 60–79
Critical >= 80
prediction:
predicted_failure_within_30d = 1 when risk_score >= 70
""".strip(),
language="text",
)
st.subheader("3) After algorithm: enriched output")
if algorithm_output_df.empty:
st.warning("Algorithm output is empty because no raw telemetry was loaded.")
else:
c5, c6, c7, c8 = st.columns(4)
c5.metric("Raw rows processed", f"{len(raw_telemetry_df):,}")
c6.metric("Algorithm output rows", f"{len(algorithm_output_df):,}")
c7.metric("Assets scored", f"{algorithm_output_df['asset_id'].nunique():,}" if "asset_id" in algorithm_output_df.columns else "0")
c8.metric("Predicted 30-day failures", f"{int(algorithm_output_df.get('algorithm_predicted_failure_within_30d', pd.Series(dtype=int)).sum()):,}")
c9, c10 = st.columns(2)
with c9:
risk_counts = algorithm_output_df["algorithm_risk_band"].value_counts().reindex(["Low", "Watch", "High", "Critical"]).dropna().reset_index()
risk_counts.columns = ["risk_band", "row_count"]
fig = px.bar(risk_counts, x="risk_band", y="row_count", title="After algorithm: risk-band distribution")
st.plotly_chart(fig, use_container_width=True)
with c10:
if {"region", "algorithm_predicted_failure_within_30d"}.issubset(algorithm_output_df.columns):
region_risk = algorithm_output_df.groupby("region", as_index=False)["algorithm_predicted_failure_within_30d"].sum()
fig = px.bar(region_risk, x="region", y="algorithm_predicted_failure_within_30d", title="Predicted 30-day failures by region")
st.plotly_chart(fig, use_container_width=True)
st.markdown("**Algorithm output preview**")
output_cols = [c for c in ["asset_id", "timestamp", "model", "region", "criticality", "days_since_last_pm", "anomaly_score", "algorithm_failure_risk_score", "algorithm_risk_band", "algorithm_predicted_failure_within_30d", "recommended_action"] if c in algorithm_output_df.columns]
st.dataframe(algorithm_output_df[output_cols].head(2000), use_container_width=True)
st.markdown("**Latest asset-level output after algorithm**")
st.dataframe(algorithm_asset_summary_df.head(500), use_container_width=True)
st.download_button(
"Download algorithm_output.csv",
data=algorithm_output_df.to_csv(index=False),
file_name="algorithm_output.csv",
mime="text/csv",
)
st.download_button(
"Download latest_asset_risk_summary.csv",
data=algorithm_asset_summary_df.to_csv(index=False),
file_name="latest_asset_risk_summary.csv",
mime="text/csv",
)
st.subheader("4) WSL commands to run dashboard")
st.code(
"""
cd "local-user-path-hidden/OneDrive - Pearce Services, LLC/onedrive_ubuntu/project/Predictive_Preventive_Maintenance_for_Generator_Reliability"
pip install -r requirements.txt
streamlit run app.py
""".strip(),
language="bash",
)
# Executive KPIs tab
# -----------------------------
with tabs[2]:
st.header("Executive KPIs")
valid_days = pd.Series(dtype=float)
if not linked_df.empty and "days_to_next_failure" in linked_df.columns:
valid_days = pd.to_numeric(linked_df["days_to_next_failure"], errors="coerce").dropna()
avg_days = valid_days.mean() if not valid_days.empty else np.nan
median_days = valid_days.median() if not valid_days.empty else np.nan
fail_30 = (valid_days <= 30).mean() if not valid_days.empty else np.nan
fail_90 = (valid_days <= 90).mean() if not valid_days.empty else np.nan
total_downtime = pd.to_numeric(failure_df.get("downtime_hours", pd.Series(dtype=float)), errors="coerce").sum()
total_failure_cost = pd.to_numeric(failure_df.get("total_cost", pd.Series(dtype=float)), errors="coerce").sum()
total_revenue_loss = 0
total_customers = 0
if not business_df.empty:
total_revenue_loss = pd.to_numeric(business_df.get("estimated_revenue_loss", pd.Series(dtype=float)), errors="coerce").sum()
total_customers = pd.to_numeric(business_df.get("estimated_customers_impacted", pd.Series(dtype=float)), errors="coerce").sum()
c1, c2, c3, c4 = st.columns(4)
c1.metric("Avg days PM → failure", num(avg_days, 1))
c2.metric("Median days PM → failure", num(median_days, 1))
c3.metric("Failure within 30 days", pct(fail_30))
c4.metric("Failure within 90 days", pct(fail_90))
c5, c6, c7, c8 = st.columns(4)
c5.metric("Total downtime hours", num(total_downtime, 1))
c6.metric("Failure repair cost", money(total_failure_cost))
c7.metric("Estimated revenue loss", money(total_revenue_loss))
c8.metric("Customer impact count", num(total_customers, 0))
st.divider()
st.subheader("Executive interpretation")
st.markdown(
f"""
- The loaded data contains **{len(asset_df):,} assets**, **{len(pm_df):,} PM events**, and **{len(failure_df):,} failure/repair tickets**.
- The average observed time from PM to the next failure is **{num(avg_days, 1)} days**.
- **{pct(fail_30)}** of linked PM events are followed by a failure within 30 days.
- The failure records represent approximately **{num(total_downtime, 1)} downtime hours** and **{money(total_failure_cost)}** in repair cost.
"""
)
if not linked_df.empty and "failure_found_flag" in linked_df.columns:
status_counts = linked_df["failure_found_flag"].value_counts().rename(index={0: "No later failure observed", 1: "Later failure observed"})
fig = px.pie(
values=status_counts.values,
names=status_counts.index,
title="Linked PM events with later failure observed",
hole=0.35,
)
st.plotly_chart(fig, use_container_width=True)
# -----------------------------
# PM Effectiveness tab
# -----------------------------
with tabs[3]:
st.header("Preventive Maintenance Effectiveness")
if linked_df.empty or "days_to_next_failure" not in linked_df.columns:
st.warning("No PM-to-failure linked data found. Upload PM and failure CSVs, then click 'Rebuild PM → Failure Links'.")
else:
linked_plot = linked_df.copy()
linked_plot["days_to_next_failure"] = pd.to_numeric(linked_plot["days_to_next_failure"], errors="coerce")
c1, c2 = st.columns(2)
with c1:
fig = px.histogram(
linked_plot.dropna(subset=["days_to_next_failure"]),
x="days_to_next_failure",
nbins=50,
title="Distribution: Days from PM to next failure",
labels={"days_to_next_failure": "Days to next failure"},
)
st.plotly_chart(fig, use_container_width=True)
with c2:
surv = survival_curve(linked_plot)
if not surv.empty:
fig = px.line(
surv,
x="days",
y="failure_free_probability",
markers=True,
title="Failure-free probability after PM",
labels={"days": "Days after PM", "failure_free_probability": "Failure-free probability"},
)
fig.update_yaxes(tickformat=".0%")
st.plotly_chart(fig, use_container_width=True)
c3, c4 = st.columns(2)
with c3:
if "ontime_flag" in linked_plot.columns:
temp = linked_plot.dropna(subset=["ontime_flag", "days_to_next_failure"]).copy()
if not temp.empty:
temp["PM status"] = temp["ontime_flag"].map({1: "On-time PM", 0: "Delayed PM"}).fillna("Unknown")
fig = px.box(
temp,
x="PM status",
y="days_to_next_failure",
points="outliers",
title="On-time vs delayed PM: Days to next failure",
)
st.plotly_chart(fig, use_container_width=True)
with c4:
if "delay_days" in linked_plot.columns:
temp = linked_plot.dropna(subset=["delay_days", "days_to_next_failure"]).copy()
if not temp.empty:
fig = px.scatter(
temp.sample(min(len(temp), 5000), random_state=42),
x="delay_days",
y="days_to_next_failure",
trendline="ols",
title="PM delay vs days to next failure",
labels={"delay_days": "PM delay days", "days_to_next_failure": "Days to next failure"},
)
st.plotly_chart(fig, use_container_width=True)
st.subheader("PM effectiveness data")
st.dataframe(linked_plot.head(1000), use_container_width=True)
# -----------------------------
# Failures & Cost tab
# -----------------------------
with tabs[4]:
st.header("Failure Patterns, Cost, and Downtime")
if failure_df.empty:
st.warning("No failure_events.csv data found.")
else:
c1, c2 = st.columns(2)
with c1:
if "failure_category" in failure_df.columns:
cat = failure_df["failure_category"].fillna("Unknown").value_counts().reset_index()
cat.columns = ["failure_category", "count"]
fig = px.bar(cat, x="failure_category", y="count", title="Failures by category")
st.plotly_chart(fig, use_container_width=True)
with c2:
if "severity" in failure_df.columns:
sev = failure_df["severity"].fillna("Unknown").value_counts().reset_index()
sev.columns = ["severity", "count"]
fig = px.bar(sev, x="severity", y="count", title="Failures by severity")
st.plotly_chart(fig, use_container_width=True)
c3, c4 = st.columns(2)
with c3:
if {"region", "total_cost"}.issubset(failure_df.columns):
temp = failure_df.copy()
temp["total_cost"] = pd.to_numeric(temp["total_cost"], errors="coerce")
reg = temp.groupby("region", as_index=False)["total_cost"].sum()
fig = px.bar(reg, x="region", y="total_cost", title="Repair cost by region")
st.plotly_chart(fig, use_container_width=True)
with c4:
if {"model", "downtime_hours"}.issubset(failure_df.columns):
temp = failure_df.copy()
temp["downtime_hours"] = pd.to_numeric(temp["downtime_hours"], errors="coerce")
mod = temp.groupby("model", as_index=False)["downtime_hours"].sum().sort_values("downtime_hours", ascending=False)
fig = px.bar(mod, x="model", y="downtime_hours", title="Downtime hours by model")
st.plotly_chart(fig, use_container_width=True)
if not business_df.empty:
st.subheader("Business impact")
b1, b2, b3 = st.columns(3)
b1.metric("Truck-roll cost", money(pd.to_numeric(business_df.get("truck_roll_cost", pd.Series(dtype=float)), errors="coerce").sum()))
b2.metric("Revenue loss", money(pd.to_numeric(business_df.get("estimated_revenue_loss", pd.Series(dtype=float)), errors="coerce").sum()))
b3.metric("SLA breach events", num(pd.to_numeric(business_df.get("sla_breach_flag", pd.Series(dtype=float)), errors="coerce").sum(), 0))
if {"event_date", "estimated_revenue_loss"}.issubset(business_df.columns):
temp = business_df.copy()
temp["event_month"] = pd.to_datetime(temp["event_date"], errors="coerce").dt.to_period("M").astype(str)
temp["estimated_revenue_loss"] = pd.to_numeric(temp["estimated_revenue_loss"], errors="coerce")
month = temp.groupby("event_month", as_index=False)["estimated_revenue_loss"].sum()
fig = px.line(month, x="event_month", y="estimated_revenue_loss", markers=True, title="Estimated revenue loss over time")
st.plotly_chart(fig, use_container_width=True)
# -----------------------------
# Telemetry Risk tab
# -----------------------------
with tabs[5]:
st.header("Telemetry and Failure Risk")
if telemetry_df.empty:
st.warning("No telemetry_weekly.csv data found.")
else:
tele = telemetry_df.copy()
for col in ["anomaly_score", "days_since_last_pm", "days_to_next_failure", "failure_within_30d"]:
if col in tele.columns:
tele[col] = pd.to_numeric(tele[col], errors="coerce")
high_risk = pd.DataFrame()
if "anomaly_score" in tele.columns:
latest = tele.sort_values("timestamp").groupby("asset_id", as_index=False).tail(1)
high_risk = latest.sort_values("anomaly_score", ascending=False).head(25)
c1, c2, c3 = st.columns(3)
c1.metric("Latest high-risk assets shown", f"{len(high_risk):,}")
c2.metric("Avg anomaly score", num(tele["anomaly_score"].mean(), 1))
if "failure_within_30d" in tele.columns:
c3.metric("Rows labeled failure within 30d", f"{int(tele['failure_within_30d'].sum()):,}")
c1, c2 = st.columns(2)
with c1:
if {"days_since_last_pm", "anomaly_score"}.issubset(tele.columns):
sample = tele.dropna(subset=["days_since_last_pm", "anomaly_score"])
sample = sample.sample(min(len(sample), sample_limit), random_state=42) if len(sample) > sample_limit else sample
color_col = "failure_within_30d" if "failure_within_30d" in sample.columns else None
fig = px.scatter(
sample,
x="days_since_last_pm",
y="anomaly_score",
color=color_col,
hover_data=["asset_id"] if "asset_id" in sample.columns else None,
title="Anomaly score vs days since last PM",
)
st.plotly_chart(fig, use_container_width=True)
with c2:
if {"days_since_last_pm", "anomaly_score"}.issubset(tele.columns):
temp = tele.dropna(subset=["days_since_last_pm", "anomaly_score"]).copy()
temp["pm_age_bucket"] = pd.cut(temp["days_since_last_pm"], bins=[0, 30, 60, 90, 120, 180, 365, 10000])
bucket = temp.groupby("pm_age_bucket", observed=True, as_index=False)["anomaly_score"].mean()
bucket["pm_age_bucket"] = bucket["pm_age_bucket"].astype(str)
fig = px.bar(bucket, x="pm_age_bucket", y="anomaly_score", title="Average anomaly score by PM age bucket")
st.plotly_chart(fig, use_container_width=True)
st.subheader("Top risky assets by latest anomaly score")
if not high_risk.empty:
st.dataframe(high_risk, use_container_width=True)
# -----------------------------
# Predictive ML & PM Strategy tab
# -----------------------------
with tabs[6]:
st.header("Predictive ML and Prescriptive PM Strategy")
st.markdown(
"""
This is the **machine-learning section** of the dashboard. It trains real supervised ML models on telemetry records,
compares their performance, scores each asset, and converts the score into maintenance actions.
**Three-phase flow:**
1. **Phase 1 - Reliability analytics:** PM completed → next failure → days to next failure.
2. **Phase 2 - Predictive ML:** telemetry + PM recency → failure probability within 14 or 30 days.
3. **Phase 3 - Prescriptive PM:** failure probability + business impact → recommended maintenance action.
"""
)
# Prefer full raw telemetry with model/region/criticality if available. Otherwise use telemetry_weekly.
ml_source_df = raw_telemetry_df.copy() if not raw_telemetry_df.empty else telemetry_df.copy()
if ml_source_df.empty:
st.warning("No telemetry dataset found for ML training.")
else:
c1, c2, c3 = st.columns(3)
target_col = c1.selectbox("ML prediction target", ["failure_within_30d", "failure_within_14d"], index=0)
max_rows = c2.slider("Training sample size", min_value=20000, max_value=120000, value=90000, step=10000)
c3.metric("ML source rows", f"{len(ml_source_df):,}")
st.subheader("ML inputs")
st.markdown(
"""
The model uses telemetry and asset-context features including:
`days_since_last_pm`, `runtime_hours_week`, `avg_load_pct`, `oil_temp_c`, `coolant_temp_c`,
`battery_voltage`, `vibration_mm_s`, `fuel_rate_lph`, `alarm_count`, `anomaly_score`,
`model`, `region`, `criticality`, and `environment_type`.
"""
)
ml_frame, numeric_cols, categorical_cols = prepare_ml_frame(ml_source_df, target_col=target_col, max_rows=max_rows)
if ml_frame.empty:
st.error(f"Cannot train ML model because `{target_col}` is not available in the telemetry dataset.")
else:
with st.spinner("Training Logistic Regression and Random Forest models..."):
ml_results = train_failure_models(ml_frame, numeric_cols, categorical_cols, target_col=target_col)
perf_df = pd.DataFrame([
{
"Model": model_name,
"Accuracy": result["accuracy"],
"Precision": result["precision"],
"Recall": result["recall"],
"F1": result["f1"],
"ROC AUC": result["roc_auc"],
}
for model_name, result in ml_results.items()
]).sort_values("ROC AUC", ascending=False)
st.subheader("Phase 2 output: trained ML model comparison")
st.dataframe(
perf_df.style.format({"Accuracy": "{:.3f}", "Precision": "{:.3f}", "Recall": "{:.3f}", "F1": "{:.3f}", "ROC AUC": "{:.3f}"}),
use_container_width=True,
)
best_model_name = perf_df.iloc[0]["Model"]
best_model = ml_results[best_model_name]["pipeline"]
st.success(f"Best model for the current data/filter selection: **{best_model_name}**")
left, right = st.columns(2)
with left:
st.subheader("Confusion matrix")
cm = ml_results[best_model_name]["confusion_matrix"]
cm_df = pd.DataFrame(cm, index=["Actual 0", "Actual 1"], columns=["Predicted 0", "Predicted 1"])
fig = px.imshow(cm_df, text_auto=True, title=f"Confusion matrix - {best_model_name}")
st.plotly_chart(fig, use_container_width=True)
with right:
st.subheader("Top predictive drivers")
fi = get_ml_feature_importance(best_model, best_model_name, top_n=15)
fig = px.bar(fi.sort_values("importance"), x="importance", y="feature", orientation="h", title="Feature importance")
st.plotly_chart(fig, use_container_width=True)
st.subheader("Phase 2 output: asset-level ML failure risk")
latest_risk = score_latest_asset_risk(ml_source_df, best_model)
if latest_risk.empty:
st.info("No latest asset risk records available.")
else:
st.dataframe(
latest_risk.head(30).style.format({"ml_predicted_failure_risk": "{:.2%}"}),
use_container_width=True,
)
if "ml_risk_band" in latest_risk.columns:
band = latest_risk["ml_risk_band"].astype(str).value_counts().reset_index()
band.columns = ["risk_band", "asset_count"]
fig = px.bar(band, x="risk_band", y="asset_count", title="Asset count by ML risk band")
st.plotly_chart(fig, use_container_width=True)
st.subheader("Phase 3 output: prescriptive PM action plan")
action_plan = build_prescriptive_pm_plan(latest_risk, linked_df, business_df)
if action_plan.empty:
st.info("No action plan could be generated.")
else:
p1, p2, p3 = st.columns(3)
p1.metric("Critical assets", f"{int((action_plan['ml_predicted_failure_risk'] >= 0.8).sum()):,}")
p2.metric("High-risk assets", f"{int(((action_plan['ml_predicted_failure_risk'] >= 0.6) & (action_plan['ml_predicted_failure_risk'] < 0.8)).sum()):,}")
p3.metric("Top-20 cost exposure", money(action_plan.head(20).get("estimated_failure_cost_exposure", pd.Series(dtype=float)).sum()))
st.dataframe(
action_plan.head(30).style.format({
"ml_predicted_failure_risk": "{:.2%}",
"estimated_failure_cost_exposure": "${:,.0f}",
"historical_median_days_to_failure": "{:.0f}",
}),
use_container_width=True,
)
if "prescriptive_pm_action" in action_plan.columns:
actions = action_plan["prescriptive_pm_action"].value_counts().reset_index()
actions.columns = ["action", "asset_count"]
fig = px.bar(actions, x="asset_count", y="action", orientation="h", title="Recommended PM actions from ML risk")
st.plotly_chart(fig, use_container_width=True)
st.subheader("What this ML proves")
st.markdown(
"""
- The dashboard is no longer only rule-based. It now **trains supervised ML models** on telemetry labels.
- Logistic Regression provides an explainable baseline.
- Random Forest captures nonlinear failure patterns from vibration, battery voltage, alarms, anomaly score, and PM recency.
- The output is not just a chart; it is an asset-level probability of near-term failure and a PM recommendation.
"""
)
# -----------------------------
# Saved Outputs tab
# -----------------------------
with tabs[7]:
st.header("Saved Outputs Included in Package")
st.markdown("These are pre-generated figures and tables from the notebook workflow.")
fig_files = sorted([p for p in OUTPUT_FIG_DIR.glob("*.png")]) if OUTPUT_FIG_DIR.exists() else []
if fig_files:
for p in fig_files:
st.subheader(p.stem.replace("_", " ").title())
st.image(str(p), use_container_width=True)
else:
st.info("No saved PNG figures found in outputs/figures.")
st.divider()
table_files = sorted([p for p in OUTPUT_TABLE_DIR.glob("*.csv")]) if OUTPUT_TABLE_DIR.exists() else []
if table_files:
st.subheader("Saved output tables")
for p in table_files:
with st.expander(p.name):
try:
df = read_csv_safely(p)
st.dataframe(df.head(500), use_container_width=True)
st.download_button(
f"Download {p.name}",
data=df.to_csv(index=False),
file_name=p.name,
mime="text/csv",
)
except Exception as e:
st.error(f"Could not read {p.name}: {e}")
else:
st.info("No saved CSV tables found in outputs/tables.")
# -----------------------------
# Data Explorer & Export tab
# -----------------------------
with tabs[8]:
st.header("Data Explorer and Export")
table_options = {
"asset_master": asset_df,
"pm_events": pm_df,
"failure_events": failure_df,
"business_impact": business_df,
"pm_failure_linked": linked_df,
"telemetry_weekly": telemetry_df,
"raw_telemetry_before_algorithm": raw_telemetry_df,
"algorithm_output_after_algorithm": algorithm_output_df,
"latest_asset_risk_summary": algorithm_asset_summary_df,
}
selected = st.selectbox("Choose table", list(table_options.keys()))
df = table_options[selected]
if df.empty:
st.warning(f"{selected} is empty.")
else:
st.write(f"Rows: **{len(df):,}** | Columns: **{len(df.columns):,}**")
st.dataframe(df.head(5000), use_container_width=True)
st.download_button(
f"Download filtered {selected}.csv",
data=df.to_csv(index=False),
file_name=f"{selected}_filtered.csv",
mime="text/csv",
)
st.divider()
st.subheader("Download all currently filtered dashboard tables")
zip_bytes = make_download_zip(table_options)
st.download_button(
"Download filtered dashboard data ZIP",
data=zip_bytes,
file_name="filtered_generator_pm_dashboard_data.zip",
mime="application/zip",
)
st.subheader("Data quality checklist")
checks = []
checks.append(("PM table has asset_id", "asset_id" in pm_df.columns if not pm_df.empty else False))
checks.append(("PM table has pm_date", "pm_date" in pm_df.columns if not pm_df.empty else False))
checks.append(("Failure table has asset_id", "asset_id" in failure_df.columns if not failure_df.empty else False))
checks.append(("Failure table has failure_date", "failure_date" in failure_df.columns if not failure_df.empty else False))
checks.append(("Linked table available", not linked_df.empty))
checks.append(("Telemetry table available", not telemetry_df.empty))
checks.append(("Business impact table available", not business_df.empty))
check_df = pd.DataFrame(checks, columns=["Check", "Passed"])
st.dataframe(check_df, use_container_width=True)
|