Preventive Maintenance

Square

This report is updated with another in its with better performance for Text processing.

Samples of maintenance Free Form English text record with start and end dates:

{” 8/23/04″, ” 8/23/04″, “TEMPER EXHAUST FAN WAS NOT WORKING. REPLACED THE FAN HUB WITH NEW BEARINGS. ALSO CHANGED THE V-BELT SIZE AP 41“}

{” 7/31/17″, ” 7/29/17″, “OSCILATOR TUBE CHANGED BY A REFURBRISHED ONE.“}

These are rows in excel files, CDN system converts the dates into integers for date i.e. number of days from the first date of maintenance history and coverts the English text into keywords and deleting the stopwords e.g. “the” or “is”:

{2, 2, “TEMPER EXHAUST FAN WORKING REPLACED FAN HUB NEW BEARINGS  CHANGED V-BELT SIZE AP 41“}

{4727, 4725, “OSCILATOR TUBE CHANGED REFURBRISHED“}

The  said textual words are then turned into numerical vectors of fixed length by TFIDF algorithm:

In information retrieval, tf–idf or TFIDF, short for term frequency–inverse document frequency, is a numerical statistic that is intended to reflect how important a word is to a document in a collection or corpus.

These vectors then increase the number of columns for the data from a few to 100s to 1000s  to 10000s depending on the usage.

Equipment #21

Words-Cloud visualization:

Note: Missing dates are formatted as ” / /” and missing reports as “” Null string in COGZ.

These vectors are then computed into Community Graphs :

For the sake of brevity this is the largest community of the Equipment #21.

Each vertex or node is a maintenance record and if you run your mouse over the node the original technician English text pops:

For the actual Web Cloud Object click on the image:

Predict Functions

Ensemble of algorithms are computed for predicting the maintenance date and the servers select the most accurate ones and report to the Cloud:

The best algorithm for Equipment #21 was computed to be:

LinearRegression: ±18.8556 days error

Note: To make the prediction more accurate the Δ days predicted for the next upcoming maintenance rather than the absolute date.

Equipment #31

Words-Cloud visualization:

Community Graphs

For the actual Web Cloud Object click on the image:

Predict Functions

The best algorithm for Equipment #21 was computed to be:

LinearRegression: ±13.5445 days error

Note: To make the prediction more accurate the Δ days predicted for the next upcoming maintenance rather than the absolute date.

Equipment #52

Words-Cloud visualization:

Community Graphs

For the actual Web Cloud Object click on the image:

Predict Functions

he best algorithm for Equipment #21 was computed to be:

NearestNeighbors: ±20.7334 days error

Note: To make the prediction more accurate the Δ days predicted for the next upcoming maintenance rather than the absolute date.