Skip to content

Production and OEE¤

How to get machine states, downtime, OEE, shift figures and part traceability from run-state, counter and reason signals. The classes live in ts_shape.events.production; each has its own page in the module reference.

Example data¤

One line over 24 hours, sampled every minute: a boolean run state, a cumulative part counter, a cumulative reject (NOK) counter, the good-part counter and a stop-reason string.

import numpy as np
import pandas as pd

rng = np.random.default_rng(1)
t = pd.date_range("2024-06-03 06:00", periods=24 * 60, freq="1min", tz="Europe/Berlin")
running = rng.random(len(t)) > 0.02                      # a short stop now and then
running[600:660] = False                                 # one long stop
parts = np.cumsum(running * 2)                           # 2 parts per running minute
nok = np.cumsum(running & (rng.random(len(t)) < 0.03))   # about 3 % rejects
reason = np.where(running, "", rng.choice(["material", "tooling"], len(t)))

def signal(uuid, column, values):
    return pd.DataFrame({"systime": t, "uuid": uuid, column: values})

df = pd.concat([
    signal("line1:running", "value_bool", running),
    signal("line1:parts", "value_integer", parts),
    signal("line1:nok", "value_integer", nok),
    signal("line1:ok", "value_integer", parts - nok),
    signal("line1:stop_reason", "value_string", reason),
], ignore_index=True)

Counters may reset (shift start, PLC restart, rollover). All counter-based methods handle that by default (handle_resets=True) and report the number of resets in a resets column.

Machine states and downtime¤

from ts_shape.events.production.machine_state import MachineStateEvents
from ts_shape.events.production.downtime_tracking import DowntimeEvents
from ts_shape.events.production.micro_stop_detection import MicroStopEvents

intervals = MachineStateEvents(df, run_state_uuid="line1:running").detect_run_idle(min_duration="5min")

downtime = DowntimeEvents(df)
by_shift = downtime.downtime_by_shift("line1:running", running_value=True)
by_reason = downtime.downtime_by_reason(
    "line1:running", "line1:stop_reason", stopped_value=False, value_column_state="value_bool"
)

micro_stops = MicroStopEvents(df, run_state_uuid="line1:running").detect_micro_stops(max_duration="3min")
  • detect_run_idle returns one interval per run or idle period; intervals shorter than min_duration are dropped.
  • downtime_by_shift returns downtime, uptime and availability_pct per shift. The run state can be a boolean (running_value=True) or a string state in value_string (running_value="Running").
  • downtime_by_reason is a Pareto of stop reasons by occurrences and duration.

Shifts default to three 8-hour shifts starting at 06:00 (shift_1 to shift_3). Pass shift_definitions={"early": ("06:00", "14:00"), ...} to the constructor of the shift-based classes to change them. A night shift that crosses midnight is counted on the day it starts (production_day=True).

OEE¤

from ts_shape.events.production.oee_calculator import OEEEvents

oee = OEEEvents(df).oee(
    "line1:running",
    "line1:parts",
    ideal_cycle_time=25.0,          # seconds per part at full speed
    total_uuid="line1:parts",
    reject_uuid="line1:nok",
)
oee[["start", "availability_pct", "performance_pct", "quality_pct", "oee_pct"]]

One row per calendar day. Availability is run time over planned time, performance is actual over ideal parts in the run time, quality is good over total parts. The basis of each percentage is in the same row (run_seconds, planned_seconds, ideal_parts, good_parts, ...). Pass planned_time_hours for a fixed planned time per day, and max_gap so that a logging gap is not booked as run time. availability, performance and quality are also available on their own.

Parts, quality and shift reports¤

from ts_shape.events.production.shift_reporting import ShiftReportEvents
from ts_shape.events.production.quality_tracking import QualityEvents
from ts_shape.events.production.shift_handover import ShiftHandoverEvents

production = ShiftReportEvents(df).shift_production("line1:parts")
quality = QualityEvents(df).nok_by_shift("line1:ok", "line1:nok")

report = ShiftHandoverEvents(df).generate_report(
    "line1:parts", "line1:ok", "line1:nok", "line1:running",
    running_value=True,
    targets={"shift_1": 900, "shift_2": 900, "shift_3": 900},
)

generate_report combines production against target, quality and availability per shift in one table. highlight_issues lists every shift metric below its threshold (by default 95 % of the production target, 98 % quality, 90 % availability).

Which class answers which question¤

Question Class Page
How many parts per hour or per part number? PartProductionEvents, LineThroughputEvents Part tracking, Line throughput
How long are the cycles, which ones are slow? CycleTimeEvents Cycle time
Where is speed lost against the target cycle time? PerformanceLossEvents Performance loss
Which stops are long, which are micro stops? LongDowntimeEvents, MicroStopEvents Long downtime, Micro stops
Which station limits the line? BottleneckDetectionEvents, LineBalancingEvents Bottleneck, Line balancing
Is a station blocked or starved? FlowConstraintEvents Flow constraints
WIP, throughput and lead time (Little's law)? FlowMetricsEvents Flow metrics
When are product changes, how long are setups? ChangeoverEvents, SetupTimeEvents Changeover, Setup time
Scrap, rework, NOK by reason and cost? ScrapEvents, ReworkEvents, QualityEvents Scrap, Rework, Quality tracking
Actual against target per shift or day? TargetEvents Target tracking
Actual against a production schedule or work orders? ProductionScheduleAdherenceEvents Schedule adherence
Output and quality per operator? OperatorPerformanceEvents Operator performance
Weekly and monthly roll-ups? PeriodSummaryEvents Period summary
Batches, their durations and yield? BatchTrackingEvents Batch tracking
Alarm floods, chattering and standing alarms? AlarmManagementEvents Alarm management
How often does a pump or valve cycle? DutyCycleEvents Duty cycle
Operating hours and starts? RuntimeAccountingEvents Runtime accounting
Machine states without a run/idle flag, from current or power? OperatingStateEvents Operating state

Traceability¤

Three classes follow parts through stations, depending on how the part id is recorded:

Class Signals
ValueTraceabilityEvents one part-id string signal per station
RoutingTraceabilityEvents one part-id signal plus an integer routing-step signal
MultiProcessTraceabilityEvents one part-id signal per process, optional handover signals between parallel cells 
from ts_shape.events.production.order_traceability import ValueTraceabilityEvents

t = pd.date_range("2024-06-03 06:00", periods=5, freq="10min", tz="UTC")
trace_df = pd.DataFrame({
    "systime": list(t[:3]) + list(t[2:5]),
    "uuid": ["press:part_id"] * 3 + ["weld:part_id"] * 3,
    "value_string": ["P-001", "P-002", "P-003"] * 2,
})
trace = ValueTraceabilityEvents(trace_df, station_uuids={"press:part_id": "Press", "weld:part_id": "Weld"})
timeline = trace.build_timeline()   # one row per part and station visit
lead_times = trace.lead_time()      # first station to last station, per part

With one sample per visit the dwell time is 0; build_timeline(infer_dwell_from_next=True) estimates it from the next sample. See Value traceability, Routing traceability and Multi-process traceability.