Andrea Martorana Tusa: Failure Prediction For Manufacturing Industry

Andrea Martorana Tusa
#sqlsat
Failure prediction for

manufacturing industry
Speaker info
• Microsoft MVP Data Platform
• Team Leader for Business Intelligence in Würth-Phoenix (Bolzano,
Italy)
• Previously worked by Widex a danish company which manufactures
hearing aids, as BI Specialist, and as BI Developer in an italian bank
• Speaker at SQL Saturdays, and other community-driven events in
Europe, (MS Cloud Summit, SQL Konferenz, SQL Nexus, SQL Days,
Dataminds Connect ...). Speaker in webinars for PASS Italian VC,
DW/BI VC
• Author for sqlservercentral.com, sqlshack.com, UGISS (User Group
Italiano SQL Server)
Agenda
• R in SQL Server 2016
• R for predictive maintenance
• Predicting failure. Build predictive models
• Scoring
• Output of R in SSMS, SSRS
• R in Power BI for data analysis and visualization
SQL Server Machine Learning Services
SQL Server Machine Learning Services is an engine
embedded into a SQL Server database, to deliver
advanced analytics calculations and perfom data
science operations.
SQL Server Machine Learning Services is available
from SQL Server 2017. Into the 2016 release it has
been introduced the capability of executing R scripts.
From 2017 release SQL Server can execute Python
code too.
R
R is the most common open source
programming language for data analysis and
statistics.
R in SQL Server
Available from 2016 release.
Two flavors. Options available during installation:
R Services (In-Database) integrated directly into SQL Server
engine.
Microsoft R Server a standalone R server installed alongside with
the database but externally.
In addition there is a community edition Microsoft R

Open and a community client Microsoft R Client.
R in SQL Server
Scalability and performance
Any R script is executed by calling a stored procedure, and the
outputs are returned as tabular results directly to SQL Server.
Performance optimization is based on scalability and
parallelization.
Two packages especially are available for handling resource
governance inside SQL Server:
• RevoScaleR a set of functions for processing, analyzing,
querying data with fast perfomances and scalability.
• RevoPemarR for writing custom parallel algorithms.
R in SQL Server
Execute R script within SQL Server
R scripts are invoked by a new system stored procedure
sp_execute_external_script to embed R (or Python)
code into a format suitable for SQL Server. The stored procedure
executes the scripts and retrieves the output.
Some parameters are passed to the sp in order to define,
language, context, input data, etc …
R in SQL Server
EXEC sp_execute_external_script
@language =N'R', Language declaration
@script=N'OutputDataSet<-InputDataSet', R script
@input_data_1 =N'SELECT 1 AS hello' Input data
WITH RESULT SETS (([hello] int not null)); Resultset from R
GO
The above R script generates the following outcome
R in SQL Server
EXEC sp_execute_external_script
@language =N'R',
@script=N'OutputDataSet<-InputDataSet',
@input_data_1 =N'SELECT 1 AS hello'
WITH RESULT SETS (([hello] int not null));
N.B.! All the output from R to SQL Server, must return the data as
a data.frame!
Any other type of object that you generate in your script must be converted to
a data frame if you want to output it as part of the stored procedure outputs.
Predictive maintenance for manufacturing
industry
”Predictive modeling is the process of using advanced statistics
and probability algorithms to predict outcomes, based on a
pretrained and built model or function.”*
Predictive maintenance solutions can provide key performance
indicators about health state and lifecycle for products in
manifacturing industries. Some use cases could be:
• Aircraft component failure
• Wind turbine failure
• Circuit breaker failure
• And, of course ……………….. Hearing aids
* Tomaz Kastrun, Julie Koesmarno. SQL Server 2017 Machine Learning Services with R. Packt Publisher
industry
R & SQL Server can be used together for building a predictive model to
address your business needs.
For example, we want to predict the probability of failure for a bicycle
component produced by the company Adventure Works.
A predictive maintenance solution helps
reducing repair costs and increasing the
lifecycle of our cycles. These models also give
insights on correlations among data, helping
to reveal patterns or outliers in the
production process.
industry
Tipical problems addressed by a predictive
maintenance solution:
• Predict Time to Failure (TTF) for a device

• Predict which component has the highest likelihood to fail
• Predict if an asset will fail in different time windows
• …
Predictive model for R in SQL Server
A predictive model for R in SQL Server is made of the following
steps:
Prepare a
Data Data model Train the Perform scoring
collection preparation (for example model and predictions
for regression)
Visualization
output
Data collection
The common data elements for failure and predictive maintenance
calculations are:
• Failure data history
• Maintenance history
• Device conditions and usage, for example data collected by sensors
• Device technical specifications
• …
Data preparation
Before getting into any type of analytical process, usually some

operations must be performed on data in order to have it the way we
need: cleansing, normalization, pre-processing, aggregation, mapping
…
Prepare a Model
A model is based on R statistical and analytical functions to explore data and
relationships between two or more variables.
Usually the first relationship to be investigated is linear regression, i.e. the nature of
relationship between two variables. It serves to estimate the Remaining Useful Life
(RUL) of an asset.
Other common models are:
• Binary classification. The task of classifying the elements of a given set into two
groups (predicting which group each one belongs to) on the basis of
a classification rule.
• Multi-class classification. The task of classifying elements of a given set into
three or more classes.
• Receiver Operating Characteristic (ROC). A curve created by plotting the true
positive rate against the false positive rate at various threshold settings. It is used
to calculate the accuracy of a test.
Train a Model
A model must learn by data. Training a model means makes it “smart” enough to
elaborate predictions by passing datasets in the same format of the test one.
To achieve the task, tipically a percentage of the dataset is used for

training/test/validation. The ratio between test and production data varies according
to your business model, experience, data quality. It could be 60/40, 70/30, etc…
A good practice is to test more than one model and analyze the returned values to
choose the one that performs best. Once the model is set up and trained, it’s
convenient to save it as stored procedure and store the data in a table for further
reuse.
Prediction and scoring
Once the model is set up it’s possible to start with predictions.
Call a stored procedure passing parameters and get predicted values.
Values are numbers which can be expressed as score, a real number between 0 and 1.
By default, scoring values above 0.5 are labelled positive and below negative.
But the threshold can be modified according to our needs, experience, sensibility …
Demo
Linear regression model in R studio
Descriptive statistics with R in SSMS
Linear regression model with R in SSMS
Multivariate analysis
Analysis of Variance
Binary and multi class models
Plotting output in Reporting Services
Models and output in Power BI
Q&A
Thanks
References
Tomaz Kastrun, Julie Koesmarno. ”SQL Server 2017 Machine Learning Services
with R”. Packt
https://docs.microsoft.com/en-us/sql/advanced-analytics/what-is-sql-server-
machine-learning?view=sql-server-2017
https://docs.microsoft.com/en-us/azure/machine-learning/team-data-science-
process/cortana-analytics-playbook-predictive-maintenance
https://gallery.azure.ai/Tutorial/Predictive-Maintenance-Template-with-SQL-
Server-R-Services-1
https://github.com/Microsoft/SQL-Server-R-Services-
Samples/blob/master/PredictiveMaintenance/SQLR/README.md

Andrea Martorana Tusa: Failure Prediction For Manufacturing Industry

Uploaded by

Copyright:

Available Formats

Andrea Martorana Tusa: Failure Prediction For Manufacturing Industry

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Andrea Martorana Tusa: Failure Prediction For Manufacturing Industry

Uploaded by

Copyright:

Available Formats

Andrea Martorana Tusa

Failure prediction for

In addition there is a community edition Microsoft R

@input_data_1 =N'SELECT 1 AS hello' Input data

WITH RESULT SETS (([hello] int not null)); Resultset from R

• Predict Time to Failure (TTF) for a device

Before getting into any type of analytical process, usually some

To achieve the task, tipically a percentage of the dataset is used for

You might also like