OperatingStateEvents

Operating states inferred from one continuous signal (spindle current, power, vibration) when the machine logs no explicit run/idle flag: state intervals, typical durations per state, unexpected state transitions and states held too long. With a boolean run signal, use MachineStateEvents instead.

ts_shape.events.production.operating_state · Production guide · API reference

Example

import numpy as np
import pandas as pd
from ts_shape.events.production.operating_state import OperatingStateEvents

# spindle current every 30 s: idle ~2 A and cutting ~18 A, alternating every 15 min
t = pd.date_range("2024-01-01 06:00", periods=480, freq="30s")
current = np.tile([2.0] * 30 + [18.0] * 30, 8)
current[240:270] = 18.0  # one cut runs on for 30 min ...
current[330:360] = 2.0   # ... and the line waits 30 min later on
rng = np.random.default_rng(0)
df = pd.DataFrame({"systime": t, "uuid": "spindle:current",
                   "value_double": current + rng.normal(0, 0.3, len(t))})

ops = OperatingStateEvents(df, "spindle:current")
states = ops.infer_states(n_states=2, min_duration="1min")
print(states[["start", "end", "state_label", "mean_value"]].head(4))
stats = ops.state_duration_statistics(n_states=2)
too_long = ops.dwell_time_violation(1, "25min")   # state 1 = "high"
print(too_long[["start", "state_label", "duration_seconds", "excess_seconds"]])
sequence = ops.unusual_state_sequence(expected_order=[0, 1, 0])

The cut that ran on is the one high interval longer than 25 minutes; it is reported with its excess over the limit.

Methods

Method	Returns
infer_states(n_states=2, min_duration='0s')	intervals: start, end, state (0 = lowest), state_label, mean_value, duration_seconds
state_duration_statistics(n_states=2)	one row per state: count, mean_duration, std_duration, min_duration, max_duration, total_duration (seconds)
unusual_state_sequence(n_states=2, expected_order=None)	one row per transition: transition_time, from_state, to_state, from_label, to_label, is_unusual
dwell_time_violation(state, max_duration, n_states=2)	intervals of state longer than max_duration: duration_seconds, excess_seconds

Notes

• States are value bins at the quantiles of the signal: with n_states=2 the split is the median. A machine that spends most of its time in one state gets that state split in two, so choose n_states (and the time span) so the states are of similar weight, or threshold the signal yourself and use MachineStateEvents.
• Labels are low/high for two states, low/mid/high for three, state_0 … otherwise. state numbers count from the lowest values.
• An interval ends at its last sample, so a phase of 15 minutes sampled every 30 s lasts 870 s.
• Without expected_order, a transition is normal when it accounts for at least 10 % of all transitions. expected_order=[0, 1, 0] allows exactly the transitions 0→1 and 1→0.
• Samples without a value are ignored; with fewer valued samples than n_states the result is empty.