Configuring Failure Curve Analytics

1. Failure Curve Analytics (FCA): A feature in APM that predicts when and how likely a technical object is to fail, based on its age.
2. Weibull Model: A statistical model commonly used in reliability engineering to represent the failure rate of assets.
3. Technical Object: The asset or component within APM to which FCA is applied.
4. Notifications: Reports or messages in the APM system that signal technical object failure, used as input data for FCA.
5. Malfunction Start Date: The date when an asset's failure began, required for training the FCA model.
6. Malfunction End Date: The date when an asset's failure ended, also required for training the FCA model.
7. Upper and Lower Confidence Intervals: Ranges in the FCA graph indicating the likely spread of failure given the technical object's age.
8. Remaining Useful Life: An estimate of the time during which a technical object is expected to perform without failure.
9. Model Configuration: The setup process for an FCA model, including selecting technical objects and defining failure modes.
10. Group: A collection of one or more technical objects used in an FCA model.
11. Age for Conditional Probability of Failure: The maximum number of days for which a technical object's failure curve is plotted.
12. Installation Date: A date that might be used to calculate the age of a technical object in relation to its failure data.
13. Technical Object is Repairable: A factor in the FCA model that influences how the age of a technical object is calculated, based on repair history.
14. Input Data Sets: The selection of technical objects, failure modes, and notifications used for training and scoring the FCA model.

In this lesson, Jessica will refine her ability to set up Failure Curve Analytics within SAP Asset Performance Management, applying AI to learn how much useful life remains in the assets managed by CRT Manufacturing.

Failure Curve Analytics (FCA) is a feature within APM that allows a user to predict when a technical object could fail and how likely failure would occur given the age of the technical object. By referencing the failure notifications associated to the technical object that signal technical object failure, we can use this to train and score an FCA model and visualize the failure data. The graph that FCA produces is based on the Weibull model, a commonly used data model within the realm of reliability. On the graph itself, we can view additional information related to the failure curve model, such as the upper and lower confidence intervals (the range in values of likely failure given the technical object's age), remaining useful life, and more.

Video Summary

Learn how to configure failure curve analytics in SAP Asset Performance Management. Follow along as Ryan guides you through the process, from creating a new configuration to training and scoring the model.

The FCA model will need notifications to train and score from. For this scenario, it is assumed that all the notifications that will be used to train the FCA model have already been created and configured properly within the S/4HANA system that is associated with the APM system. As generating the notifications and configuring the failure modes to be used with technical objects within FCA are outside the scope of this document, they will not be discussed in much more detail. To utilize a notification for APM, certain requirements must be met with these notifications. Here are those requirements listed below.

• The notifications must be assigned to the technical object(s) being worked on in the FCA training model.
• The notification can only be used if the failure mode associated with it is set to be used in the FCA model (which is done automatically).
• There should be at least 10 notifications being worked with in a model.
• The notification must have a malfunction start date.
• The notification must also have a malfunction end date that is at least one day after the start date.
• None of the malfunction dates can overlap between notifications.
• The gap between days for each successive notification must be larger as time goes on.
• Ex. The 1st notification has a malfunction start and end date of May 1st and May 2nd respectively, the second notification has a malfunction start and end date of May 3rd and May 5th, the 3rd notification has a malfunction start and end date of May 6th and May 9th, and so on.

As APM does not have any way to manage notifications or failure modes directly from the application itself, these must be handled by other means. This could be either via the S/4HANA system directly or by means of something such as APIs to create or modify this data.

To view the actual failure curve graph that is generated, we need to create an FCA model. Inside of the model is where we will specify the technical object(s) that are having failure curve analytics being done on then as well as the failure modes that we will be making the graphs on and the date range of notifications that we will be using to train the model. To create an FCA model, open the Failure Curve Analytics Configuration tile on the main page of APM.

This will open the Failure Curve Analytics page. On this page are a list of FCA models and some additional information pertaining to those models. To create a new FCA model, choose the Create button on the top right of the page next to the search bar.

A pop-up will appear once the button is choosen. In here, we will enter all the necessary information for our FCA model. For now, just enter the Model Configuration, which is essentiality the name of the FCA model being made. The Description field can also be added in if desired. A Group also needs to be either assigned or created. This group will contain one or more technical objects that are being used for the FCA model. Either choose the Create button to make a new group or the square icon to the left of the Create button to select a previously made group.

Enter a name for the group in the Group field in the new Create group pop-up. Optionally enter a Description as well as a Long Text. Once finished, choose the Save button on the bottom of the pop-up.

After having choosen the Save button, another pop up displays, that will prompt you to select one or more technical objects for the group. Once your desired technical objects are checked off, choose the OK button at the bottom left of the page. Make sure to utilize the search bar as well as the other filters on the page to narrow down the list of potential technical objects to choose from. If selecting one or more of the filter options, make sure to choose the Go button at the top right of the pop up to ensure the technical object list is being filtered out by the filter parameters just set.

With the Group either newly created or selected from the list of previous Groups, the remaining fields can optionally be entered or adjusted. Luckily, each of the remaining fields have an i icon to the right of them that gives more context as to what each field represents.

The Age for Conditional Probability of Failure entered here represents the maximum number of days that a technical object will have its failure curve plotted out to. Note that the age value given must be greater than the age of the technical object being evaluated. Most technical objects will have a starting date of zero, so this will typically be something that does not have to be worried about.

The First Installation Date is Maintained and Reliable value is used in determining the age of the technical object in your FCA graph. If yes is selected, then the installation date of the technical object will be used in determining how old the technical object is given the failure dates specified by the notifications being used in the model. If no is selected, then the age of the technical object for the first failure notification cannot be determined. On subsequent failures, the age used here will be based off the number of days since the first failure notification occurred. The installation date of a technical object is set in the SAP S/4HANA end and cannot be set within APM itself.

The Technical Object is Repairable value is also used for determining the age of a technical object when used for the FCA model. If yes is selected, the age of the technical object is based on the number of days since the last repair date of the technical object. If no is selected, the age of the technical object is based on the number of days since the installation date. The repair date of a technical object is set in S/4HANA.

With all desired values entered in, choose the Save button on the bottom right of the pop up to finalize the creation of the FCA model.

Now with the FCA model created, we can go about making some final configurations before going ahead with the training and scoring of the model. Technically, the model can immediately be trained and scored if the preset configurations are to your liking. All these configurations will be set up in the Input Data Sets tab of the FCA model page.

In the Header section, the Edit Header button can be choosen to modify information relating to the Header. Unlike when creating the FCA model however, the Model Configuration and Group fields cannot be altered. After all changes to the model are made, choose the Save button on the bottom right of the pop up.

The first section on this page is the Technical Objects section. This is where the technical object(s) that were assigned within the group are displayed. These are the technical objects that are being trained and scored within this FCA model. Additional technical objects can be put into the model and technical objects can also be removed from the model. Choose either the Add or Remove buttons on the right side of the Technical Objects section of the page to change which technical objects are in the FCA model. Note that the Remove button will be unusable if one or more technical objects aren't checked off for removal.

The next section of the page is the Failure Modes section. Within this section, all the failure modes associated with the Failure Data Profiles of the technical objects used in the FCA model are displayed. A Failure Data Profile in APM (also referred to as a Catalog Profile within the S/4HANA side) for our purposes here is essentially a grouping of failure modes that can be assigned to technical objects. These Failure Data Profiles are assigned to technical objects in S/4HANA, and their contents are also configured within S/4HANA.

Even though all the failure modes associated with the failure data profiles of the technical objects are displayed, failure modes may not be used for training and scoring the model. This would be the case if the failure notifications used for the model do not have enough of the listed failure mode assigned to them to adequately train and score the model. For example, if failure mode A is listed in 15 failure notifications but failure mode B is listed in only 5 failure notifications, then only failure mode A will have an FCA graph mapped out for it.

Even with all the potential failure modes listed here, failure modes can still be added or removed from the model if we don't wish to graph them out. Check off the failure modes to be excluded from training and scoring and choose the Remove button on the right side of the failure mode section to remove them. If you wish to add them back, choose the Add button at the same location to put them back into the model. We will keep all the failure modes here for this model though. Note that it is considered best practice to have no more than 40 failure modes associated with a model to ensure

The final section of the page is the Notification Date Ranges section. This section simply lets us specify which notifications will be used for training and scoring the FCA model given both the malfunction start date and malfunction end date are within the date range. A default date range is given, although this can be changed if desired. Multiple date ranges can also be given if you wish to skip over notifications. Select the radio button for a notification date range and choose Edit to be able to change the date ranges. Choose Save to persist the changes and cancel to revert to the original settings. Choose the Add button to have additional date ranges used for the model. Select a radio button for any additional date ranges added and choose the Remove button to delete the date range from the model.

With all the Input Data Sets properly configured, the FCA model can now be trained and scored. To do so, choose the Train and Score button at the top right of the page near the rest of the header information. Once chosen, the model will take some time to either come back with a new FCA graph or return with an error that we would need to fix.

The Train and Score button will now be partially faded, and The Operational Status, Training Status, and Scoring Status will now be set to either Pending or In Progress. Moving to the Training and Scoring tabs will show the logs pertaining to the progress of each. They will also list any failures that may prevent the model from completing a successful training or scoring.

If the training and scoring of the model are both successful, the status of both will be set to Completed. If not, the status will be set to Failed. We can view the logs in the Trainings and Scorings tabs to see how far along each got within their process and what was the reason behind the failure of each if applicable. On each of these tabs, we can see the history of all the training and scoring runs that occurred within this model. The refresh button on the right side of the page in the training/scoring runs sections of their respective tabs allows you to see any additional trainings/scorings that may have occurred since loading the page.

After a successful or unsuccessful training and scoring, the option to train and score again is available. Training and scoring the model should only happen again if changes have been made to either the model itself or within the S/4HANA side. The status of the model will switch back to In Process and the result status of the new training and scoring will appear within the Scoring Runs section of the Trainings and Scorings tabs.

For reference, here is a separate FCA model with a failed training and scoring. In the Trainings Logs, it states that the reason for failure here was because there were no notifications that could be used for training the model for the technical objects being used here. The scorings failed as there were no successful trainings which occurred. The reason for failures between all the FCA models will vary case by case.

Now that the FCA model is properly trained and scored, the FCA analytics graph can be viewed within the Technical Object Details page. To navigate to this page, go back to the main page and choose the Explore Technical Objects tile. This is the first tile of the Master Data section of the page.

On the Explore Technical Objects page, choose the technical object that was used for the FCA model. Utilize the filters on the page if necessary to find the technical object more easily.

When switching over to the Analytics tab for the technical object, we can view the newly created FCA graph. We can switch the FCA graph between the different failure modes associated with the model it was based on as well as which model we want to view. Multiple models can perform FCA on the same technical object, so the ability to switch between models is present as well. If not done so already, select the FCA model that was just trained and scored and choose a failure mode that outputs an FCA graph.

On the graph, there are three curves. The main one is the Probability of Failure curve, which is the likely percent chance this technical object will fail because of the failure mode selected at a specified age in days. There are also the Upper and Lower Confidence Intervals. These mark the highest and lowest percent chance that the technical object will fail due to the specified failure mode. For example, the graph shown here states that by day 20 the chance the pump will fail due to heading will range anywhere between 64% and 98%. With that said, the mean (or average) chance calculated for this pump to fail because of heating is at 87%. This way, we can lean more towards one side of our failure range rather than the other.

Also on this page is the Predicted Failure Date and the Time to Failure fields next to the Failure Mode field. The Predicted Failure Date estimates the day in which this technical object will fail due to the failure mode set in the configuration here and the Time to Failure is the number of days in between when the assessment was last scored and the Predicted Failure Date. One thing to note is that the Predicted Failure Date will never exceed the Age for Conditional Probability of Failure that was set within the FCA model.

Configuring Failure Curve Analytics

How might the concepts of the Weibull model and confidence intervals help you in planning and decision-making for future projects or maintenance schedules?

Expert Response to Personal Reflection Question:

When managing a project, predicting issues ahead of time is like using SAP's Failure Curve Analytics. Historical data guides me to foresee and prevent problems. The Weibull model helps by showing failure probability over an asset's life, while confidence intervals reveal how sure we can be about those predictions.

These tools are like a forecast, helping me to plan maintenance or backup plans effectively. They encourage a proactive stance-preparing for what's likely to happen, not just reacting when it does. This strategic approach is key for smooth operations and successful project outcomes.

Overview:

• Failure Curve Analytics (FCA) in APM enables users to predict technical object failures based on notifications, using the Weibull model for reliability predictions.
• FCA graphs display failure curves, upper and lower confidence intervals, remaining useful life, and more.

Ensure Notifications Exist for Training and Scoring:

• Notifications required for FCA model training and scoring.
• Notifications should be properly configured in the S/4HANA system associated with APM.
• Requirements for notifications include being assigned to technical objects, specific failure modes, at least 10 notifications, valid malfunction start and end dates, and non-overlapping dates.

Creating an FCA Model:

1. Open the Failure Curve Analytics Configuration tile on the APM main page.
2. Choose Create to add a new FCA model.
3. Enter Model Configuration (name) and assign or create a group containing technical objects.
4. Configure additional fields: Age for Conditional Probability of Failure, First Installation Date, Technical Object is Repairable, etc.
5. Save to finalize the FCA model creation.

Configuring an FCA Model:

• Edit technical objects, failure modes, and notification date ranges in the Input Data Sets tab of the FCA model page.
• Specify the age for conditional probability of failure, first installation date, and repairability options.
• Save to apply configurations.

Training and Scoring the Model:

1. Choose Train and Score to initiate the training and scoring process.
2. Check the Training and Scoring tabs for progress and logs.
3. Successful training and scoring display as "Completed"; failures display as "Failed."
4. Option to train and score again if model or SAP S/4HANA changes occur.

Viewing the Analytics Graph of the FCA Model:

1. Navigate to the Explore Technical Objects tile on the main page.
2. Choose the technical object used for the FCA model.
3. Switch to the Analytics tab to view the FCA graph.
4. Select the FCA model and choose a failure mode for the graph.
5. Analyze the Probability of Failure curve, Upper and Lower Confidence Intervals, Predicted Failure Date, and Time to Failure.

Conclusion:

• FCA in APM empowers users to predict technical object failures using notifications and the Weibull model.
• Proper configuration of notifications, FCA model creation, and training/scoring processes are essential for accurate reliability predictions.
• FCA graphs provide valuable insights into the likelihood of technical object failures and remaining useful life.