Fall 2023 Work

Goals

There were 3 goals that were outlined for this project and over the course of the semester:

1. Predictive Maintenance: Forecast blade wear and anticipate blade replacement needs proactively.
2. Anomaly Detection: Identify irregularities in saw behavior during operation or over time.
3. Usage Monitoring: Monitor machine usage, track operational hours per day, determine blade lifespan, and facilitate manual blade replacement by PIs.

Approach

This is an approach that was developed in the Fall 2023 Semester to implement RUL (Remaining Useful Life) in Predictive Maintenance. The only issue was that an algorithm to detect when interval states wasn't developed until the end of the semester.

1. Data Normalization – Normalize data to a 0-1 range using the min-max algorithm for equalizing values.
2. Moving Average – Apply a moving average to smooth data points.
3. Variance Calculation – Calculate variance of the moving average to filter out outliers affecting Remaining Useful Life (RUL) calculation.
4. Derivative Calculation / Interval Generation – Derive the moving average, identify critical points, and determine trends (increasing, decreasing, constant) based on a set threshold.
5. Stratification – Group data into intervals (rest, acceleration, operation, deceleration) based on known states from the previous step.
6. Regression Calculation – Running regression on interval states, compare standard deviations, and calculate RUL by measuring distance from a predefined error threshold.

Implementation

Only Usage Monitoring was successfully implemented during Fall 2023 Semester.

1. Import all required modules & load dataset

import matplotlib.pyplot as plt
import pandas as pd

data = '/content/10-31-23-Disc-Sander.csv' # data_file 
df = pd.read_csv(data) 

time = "timestamp" # timestamp column
data = "ACC X" # data column 

2. Shift All Values by Average

shift_value = df[data].mean() 
df[data] = df[data] - shift_value 

3. Convert to Absolute Value

df[data] = df[data].abs() 

4. Normalize

def min_max(x): 
  normal_x = (x-df[data].min())/(df[data].max()-df[data].min()) 
  return normal_x 

df[data] = df[data].apply(min_max) 

5. Moving Average

df['SMA'] = df[data].rolling(20).mean() # Some value between 10 - 30 is acceptable 

6. Threshold

def threshold(x): 
  if x >= 0.015: 
    return True 
  else: 
    return False 
df["Usage"] = df["SMA"].apply(threshold) 

7. For Loop to get number of seconds – uptime of machine

print("Seconds in operation: " + str(df['Usage'].value_counts()[True] * 0.1)) 

Future Work (To Be Completed)

The Usage Monitoring based on the very limited dataset and setting has a 91% accuracy meaning it's meaning 9% of the data. Ideally, in the Spring 2024 semester, this method will be refined to have near 100% accuracy and then the algorithm for predicative maintenance can be further developed. Furthermore, the script above should be converted into a function that can get called.