name: work-sampling-analyzer description: 用于活动分布和利用率研究的工作抽样分析技能。 allowed-tools: Bash(*) Read Write Edit Glob Grep WebFetch metadata: author: babysitter-sdk version: “1.0.0” category: work-measurement backlog-id: SK-IE-035
工作抽样分析器
您是工作抽样分析器 - 专门用于工作抽样研究以分析活动分布和设备/工人利用率的技能。
概述
此技能支持AI驱动的工作抽样，包括：
- 随机观察调度
- 样本量确定
- 活动分类
- 统计置信区间
- 控制图监控
- 多活动研究
- 从抽样开发标准时间
- 利用率分析
功能
1. 样本量确定
import numpy as np
from scipy import stats
import random
from datetime import datetime, timedelta
def determine_sample_size_binomial(estimated_proportion: float,
desired_accuracy: float,
confidence_level: float = 0.95):
"""
确定工作抽样所需的样本量
estimated_proportion: 活动中估计的时间比例（十进制）
desired_accuracy: 期望精度（例如，0.05表示±5%）
confidence_level: 统计置信度（通常为0.95）
"""
p = estimated_proportion
e = desired_accuracy
z = stats.norm.ppf(1 - (1 - confidence_level) / 2)
# n = (z² × p × (1-p)) / e²
n = (z ** 2 * p * (1 - p)) / (e ** 2)
return {
"required_observations": int(np.ceil(n)),
"estimated_proportion": p,
"desired_accuracy": f"±{e * 100:.1f}%",
"confidence_level": f"{confidence_level * 100:.0f}%",
"z_score": round(z, 2)
}
def update_sample_size(observations: int, observed_proportion: float,
desired_accuracy: float, confidence_level: float = 0.95):
"""
基于观察数据更新样本量
"""
z = stats.norm.ppf(1 - (1 - confidence_level) / 2)
# 使用观察比例重新计算
p = observed_proportion
required = int(np.ceil((z ** 2 * p * (1 - p)) / (desired_accuracy ** 2)))
return {
"current_observations": observations,
"observed_proportion": round(p, 3),
"required_observations": required,
"additional_needed": max(0, required - observations),
"study_complete": observations >= required
}
2. 随机观察调度
def generate_random_schedule(study_duration_days: int,
observations_per_day: int,
work_start: str = "08:00",
work_end: str = "17:00",
exclude_lunch: tuple = ("12:00", "13:00")):
"""
为工作抽样研究生成随机观察时间
"""
start_time = datetime.strptime(work_start, "%H:%M")
end_time = datetime.strptime(work_end, "%H:%M")
lunch_start = datetime.strptime(exclude_lunch[0], "%H:%M")
lunch_end = datetime.strptime(exclude_lunch[1], "%H:%M")
schedule = []
for day in range(study_duration_days):
day_schedule = []
# 生成随机时间
attempts = 0
while len(day_schedule) < observations_per_day and attempts < 1000:
# 从开始到结束的随机分钟数
total_minutes = (end_time - start_time).seconds // 60
random_minutes = random.randint(0, total_minutes)
obs_time = start_time + timedelta(minutes=random_minutes)
# 检查是否在午餐时间
if lunch_start <= obs_time < lunch_end:
attempts += 1
continue
# 检查最小间隔（10分钟）
too_close = False
for existing in day_schedule:
if abs((obs_time - existing).seconds) < 600: # 10分钟
too_close = True
break
if not too_close:
day_schedule.append(obs_time)
attempts += 1
day_schedule.sort()
schedule.append({
'day': day + 1,
'times': [t.strftime("%H:%M") for t in day_schedule]
})
return {
"total_days": study_duration_days,
"observations_per_day": observations_per_day,
"total_observations": study_duration_days * observations_per_day,
"schedule": schedule
}
3. 活动分析
def analyze_observations(observations: list, categories: list):
"""
分析工作抽样观察结果
observations: 观察到的类别列表
categories: 可能的类别列表
"""
total = len(observations)
# 按类别计数
counts = {cat: observations.count(cat) for cat in categories}
# 计算比例和置信区间
z = stats.norm.ppf(0.975) # 95%置信度
results = []
for cat in categories:
count = counts[cat]
p = count / total if total > 0 else 0
# 标准误差
se = np.sqrt(p * (1 - p) / total) if total > 0 else 0
# 置信区间
ci_lower = max(0, p - z * se)
ci_upper = min(1, p + z * se)
results.append({
'category': cat,
'count': count,
'proportion': round(p, 4),
'percentage': round(p * 100, 1),
'std_error': round(se, 4),
'ci_95_lower': round(ci_lower * 100, 1),
'ci_95_upper': round(ci_upper * 100, 1)
})
# 按比例降序排序
results.sort(key=lambda x: x['proportion'], reverse=True)
return {
"total_observations": total,
"categories": results,
"summary": {
"productive": sum(r['proportion'] for r in results
if 'idle' not in r['category'].lower() and
'delay' not in r['category'].lower()) * 100,
"non_productive": sum(r['proportion'] for r in results
if 'idle' in r['category'].lower() or
'delay' in r['category'].lower()) * 100
}
}
4. 工作抽样控制图
def create_sampling_control_chart(daily_observations: list,
target_proportion: float = None):
"""
创建控制图以监控抽样一致性
daily_observations: 每日观察数据列表 {'day': int, 'productive': int, 'total': int}
"""
# 计算总体平均值
total_productive = sum(d['productive'] for d in daily_observations)
total_obs = sum(d['total'] for d in daily_observations)
p_bar = total_productive / total_obs if total_obs > 0 else 0
# 计算每日控制限（可变样本量）
chart_data = []
for day_data in daily_observations:
n = day_data['total']
p = day_data['productive'] / n if n > 0 else 0
# 标准误差
se = np.sqrt(p_bar * (1 - p_bar) / n) if n > 0 else 0
# 3-sigma控制限
ucl = min(1, p_bar + 3 * se)
lcl = max(0, p_bar - 3 * se)
# 检查是否在控制中
in_control = lcl <= p <= ucl
chart_data.append({
'day': day_data['day'],
'proportion': round(p, 4),
'sample_size': n,
'ucl': round(ucl, 4),
'lcl': round(lcl, 4),
'in_control': in_control
})
# 识别失控点
ooc_points = [d for d in chart_data if not d['in_control']]
return {
"center_line": round(p_bar, 4),
"chart_data": chart_data,
"out_of_control_days": len(ooc_points),
"process_stable": len(ooc_points) == 0,
"recommendation": "过程稳定" if len(ooc_points) == 0
else f"调查{len(ooc_points)}个失控点"
}
5. 从工作抽样计算标准时间
def calculate_standard_time_from_sampling(sampling_results: dict,
total_study_time_hours: float,
units_produced: int,
allowance_percent: float):
"""
从工作抽样数据开发标准时间
sampling_results: analyze_observations的结果
total_study_time_hours: 研究覆盖的总时间
units_produced: 研究期间生产的单位数量
allowance_percent: PFD津贴
"""
# 找到生产时间比例
productive_proportion = sampling_results['summary']['productive'] / 100
# 总生产时间
productive_hours = total_study_time_hours * productive_proportion
# 每单位正常时间
normal_time_hours = productive_hours / units_produced if units_produced > 0 else 0
normal_time_minutes = normal_time_hours * 60
# 应用津贴
allowance_factor = 1 + (allowance_percent / 100)
standard_time_minutes = normal_time_minutes * allowance_factor
return {
"study_duration_hours": total_study_time_hours,
"productive_proportion": round(productive_proportion, 3),
"productive_hours": round(productive_hours, 2),
"units_produced": units_produced,
"normal_time_per_unit": round(normal_time_minutes, 3),
"allowance_percent": allowance_percent,
"standard_time_per_unit": round(standard_time_minutes, 3),
"pieces_per_hour": round(60 / standard_time_minutes, 1) if standard_time_minutes > 0 else 0
}
6. 多活动研究
def multi_activity_study(observations: list, workers: list, machines: list):
"""
分析工人和机器的多活动工作抽样
observations: 观察数据列表 {'time': str, 'worker': str, 'activity': str, 'machine': str}
"""
total_obs = len(observations)
# 按工人分析
worker_analysis = {}
for worker in workers:
worker_obs = [o for o in observations if o['worker'] == worker]
n = len(worker_obs)
activities = {}
for obs in worker_obs:
act = obs['activity']
activities[act] = activities.get(act, 0) + 1
worker_analysis[worker] = {
'observations': n,
'activities': {k: {'count': v, 'percent': round(v / n * 100, 1)}
for k, v in activities.items()}
}
# 按机器分析
machine_analysis = {}
for machine in machines:
machine_obs = [o for o in observations if o.get('machine') == machine]
n = len(machine_obs)
states = {}
for obs in machine_obs:
state = obs.get('machine_state', 'running')
states[state] = states.get(state, 0) + 1
if n > 0:
machine_analysis[machine] = {
'observations': n,
'states': {k: {'count': v, 'percent': round(v / n * 100, 1)}
for k, v in states.items()},
'utilization': round(states.get('running', 0) / n * 100, 1)
}
return {
"total_observations": total_obs,
"worker_analysis": worker_analysis,
"machine_analysis": machine_analysis,
"summary": {
"avg_worker_utilization": np.mean([
100 - w['activities'].get('idle', {}).get('percent', 0)
for w in worker_analysis.values()
]),
"avg_machine_utilization": np.mean([
m['utilization'] for m in machine_analysis.values()
]) if machine_analysis else 0
}
}
流程集成
此技能与以下流程集成：
work-measurement-analysis.jsutilization-improvement.jsworkforce-planning.js
输出格式
{
"study_summary": {
"total_observations": 500,
"study_duration_days": 10,
"confidence_level": "95%"
},
"activity_analysis": {
"productive": {"percent": 72.5, "ci": [68.5, 76.5]},
"idle": {"percent": 15.2, "ci": [12.1, 18.3]},
"delay": {"percent": 12.3, "ci": [9.5, 15.1]}
},
"utilization": {
"worker": 84.8,
"machine": 78.2
},
"standard_time": {
"normal_time": 2.45,
"standard_time": 2.82
},
"recommendations": [
"通过更好的调度减少空闲时间",
"调查延迟原因"
]
}
最佳实践
- 真正随机的时间 - 使用随机数生成器
- 足够的观察次数 - 基于期望精度
- 一致的分类 - 培训所有观察员
- 简要观察 - 即时记录所见
- 告知工人 - 减少霍桑效应
- 监控稳定性 - 使用控制图
限制
- 无法确定活动顺序
- 需要清晰的类别定义
- 随机调度可能错过罕见事件
- 工人行为在被观察时可能改变