On October 7, 2026, AWS will discontinue support for Amazon Lookout for Equipment. After October 7, 2026, you will no longer be able to access the Lookout for Equipment console or resources. For more information, [see the following](https://aws.amazon.com/blogs/machine-learning/preserve-access-and-explore-alternatives-for-amazon-lookout-for-equipment/).
# Best practices with Lookout for Equipment
Training a machine learning (ML) model can involve inputs from up to 300 sensors, and you can have up to 3000 sensors represented in a single dataset. We highly recommend that you consult a subject matter expert (SME) when setting up Lookout for Equipment to monitor your equipment. This will help you get the most out of Lookout for Equipment.
We also recommend that you understand and follow the best practices described in this topic. There are three key pillars essential to setting up Lookout for Equipment for the best possible results:
+ Selecting the right application
+ Selecting the right data inputs
+ Working with SMEs to select the inputs and evaluate the results
# Choosing the right application
Choosing the right application of Lookout for Equipment involves finding the right combination of business value, equipment operations, and available data. You determine this by working directly with a subject matter experts (SME) on your equipment. Your team should consider the following:
+ **The high cost of downtime** – Equipment that can either be costly to fix or that is critical to a process is a prime candidate for monitoring.
+ **Consistency in operations** – Lookout for Equipment works best on equipment that is stationary and primarily does a continuous, stable task. A heavy duty pump that is permanently installed in a location is a good example.
+ **Relevant data** – Having data that is relevant to the critical aspects of the equipment is essential. Your equipment should have sensors that monitor these critical aspects, so that they can provide data that is relevant to how your equipment could fail. Having this data can make the difference between inference results that can effectively catch potential failures and abnormal behavior, and results that don't.
+ **Significant historical data** – Ideally, the data you use to train the machine learning (ML) model should represent all of the equipment's operating modes. For instance, when creating a model for a pump with variable speeds, the dataset should contain measurements that include an adequate amount of historical data for all of the pump speeds. For effective analysis, Lookout for Equipment should have at least six months of historical data, although a longer history is preferred. For equipment affected by seasonality, at least one year of data is highly recommended.
+ **List of historical failures (that is, labels)** – Lookout for Equipment uses data on historical failures to enhance the model's knowledge of normal equipment conditions. It looks for abnormal behavior that occurred ahead of historical failures. With more examples of historical failures, Lookout for Equipment can better develop its knowledge of healthy conditions and the unhealthy conditions that occur prior to failures. The definition of a failure can be subjective, but we have found that looking for issues that cause unplanned downtime is a good method to identify failure. For best results, give Lookout for Equipment label data for every known time period where the equipment had issues or abnormal behavior.
**Note**
Lookout for Equipment is ultimately dependent on *your* data. We cannot guarantee that there are patterns in your data that will enable Lookout for Equipment to detect failures. Determining the right set of inputs might require multiple iterations through the Lookout for Equipment model training and monitoring process. For the greatest chance of success, we highly recommend working with a subject matter expert to identify the right application and data.
# Choosing the right data
Your dataset should contain time-series data that's generated from an industrial asset such as a pump, compressor, motor, and so on. Each asset should be generating data from one or more sensors. The data that Lookout for Equipment uses for training should be representative of the condition and operation of the asset. Making sure that you have the right data is crucial. We recommend that you work with a SME. A SME can help you make sure that the data is relevant to the aspect of the asset that you're trying to analyze. We recommend that you remove unnecessary sensor data. With data from too few sensors, you might miss critical information. With data from too many sensors, your model might overfit the data and it might miss out on key patterns.
**Important**
Choosing the right input data is crucial to the success of using Lookout for Equipment. It might take multiple iterations of trial and error to find the right inputs. We cannot guarantee results. Success is highly dependent on the relevancy of your data to equipment issues.
Use these guidelines to choose the right data:
+ **Use only numerical data** – Remove nonnumerical data. Lookout for Equipment can't use non-numerical data for analysis.
+ **Use only analog data** – Use only analog data (that is, many values that vary over time). Using digital values (also known as categorical values, or values that can be only one of a limited number of options), such as valve positions or set points, can lead to inconsistent or misleading results.
+ **Remove continuously increasing data** – Remove data that is just an ever-increasing number, such as operating hours or mileage.
+ **Use data for the relevant component or subcomponent** – You can use Lookout for Equipment to monitor an entire asset (such as a pump) or just a subcomponent (such as a pump motor). Determine where your downtime issues occur and choose the component or subcomponent that has the greater effect on that.
When formatting a predictive maintenance problem, consider these guidelines:
+ **Data size** – Although Lookout for Equipment can ingest more than 50 GB of data, it can use only 7 GB with a model. Factors such as the number of sensors used, how far back in history the dataset goes, and the sample rate of the sensors can all determine how many measurements this amount of data can include. This amount of data also includes the missing data imputed by Lookout for Equipment.
+ **Missing data** – Lookout for Equipment automatically fills in missing data (known as imputing). It does this by forward filling previous sensor readings. However, if too much original data is missing, it might affect your results.
+ **Sample rate** – *Sample rate* is the interval at which the sensor readings are recorded. Use the highest frequency sample rate possible without exceeding the data size limit. The sample rate and data size might also increase your ML model training time. Lookout for Equipment handles any timestamp misalignment.
+ **Number of sensors** – Lookout for Equipment can train a model with data from up to 300 sensors. However, having the right data is more important than the quantity of data. More is not necessarily better.
+ **Vibration** – Although vibration data is usually important for identifying potential failure, Lookout for Equipment does not work with raw high-frequency data. When using high-frequency vibration data, first generate the key values from the vibration data, such as RMS and FFT.
## Filtering for normal data
Make sure that you use only data from normal (standard) operations. To do this, identify a key operating metric that indicates that the equipment is operating in a standard fashion. For example, when operating a compressor in a refinery, the key metric is usually production flow rate. In this case, you would need to filter out times when the production flow rate is below normal due to reduced production or any reason other than abnormal behavior. Other examples of key metrics might be RPM, fuel efficiency, "run" state, availability, and so on. Lookout for Equipment assumes that the data is normal. Making sure that the data fits this assumption is very important.
## Using failure labels
To provide insight into past events, Lookout for Equipment uses labels that call out these events for the ML model. Providing this data is optional, but if it's available, it can help train your model more accurately and efficiently.
For information about using labels, see [Understanding labeling](understanding-labeling.md) and [Labeling your data](labeling-data.md).
# Evaluating the output
After a model is trained, Lookout for Equipment evaluates its performance on a subset of the dataset that you've specified for evaluation purposes. It displays results that provide an overview of the performance and detailed information about the abnormal equipment behavior events and how well the model performed when detecting those.
Using the data and failure labels that you provided for training and evaluating the model, Lookout for Equipment reports how many times the model's predictions were true positives (how often the model found the equipment anomaly that was noted within the ranges shown in the labels). Within a labeled time range, the forewarning time represents the duration between the earliest time when the model found an anomaly and the end of the labeled time range.
For example, if Lookout for Equipment reports that "6/7 abnormal equipment behavior events were detected within label ranges with an average forewarning time of 32 hrs," in 6 out of the 7 labeled events, the model detected that event and averaged 32 hours of forewarning. In one case, it did not detect the event.
Lookout for Equipment also reports the abnormal behavior events that were not related to a failure, along with the duration of these abnormal behavior events. For example, if it reports that "5 abnormal equipment behavior events were detected outside the label range with an average duration of 4 hrs," the model thought an event was occurring in 5 cases. An abnormal behavior event such as this one might be attributed to someone erroneously operating the equipment for a period of time or a normal operating mode that you haven't seen previously.
Lookout for Equipment also displays this information graphically on a chart that shows the days and events and in a table.
Lookout for Equipment provides detailed information about the anomalous events that it detects. It displays a list of sensors that provided the data to indicate an anomalous event. This might help you determine which part of your asset is behaving abnormally.
# Improving your results
To improve the results, consider the following:
+ Did unrecorded maintenance events, system inefficiencies, or a new normal operating mode happen during the time of flagged anomalies in the test set? If so, the results indicate those situations. Change your train-evaluation splits so that each normal mode is captured during model training.
+ Are the sensor inputs relevant to the failure labels? In other words, is it possible that the labels are related to one component of the equipment but the sensors are monitoring a different component? If so, consider building a new model where the sensor inputs and labels are relevant to each other and drop any irrelevant sensors. Alternatively, drop the labels you're using and train the model only on the sensor data.
+ Is the label time zone the same as the sensor data time zone? If not, consider adjusting the time zone of your label data to align with sensor data time zone.
+ Is the failure label range inadequate? In other words, could there be anomalous behavior outside of the label range? This can happen for a variety of reasons, such as when the anomalous behavior was observed much earlier than the actual repair work. If so, consider adjusting the range accordingly.
+ Are there data integrity issues with your sensor data? For example, do some of the sensors become nonfunctional during the training or evaluation data? In that case, consider dropping those sensors when you run the model. Alternatively, use a training-evaluation split that filters out the non-functional part of the sensor data.
+ Does the sensor data include uninteresting normal-operating modes, such as off-periods or ramp-up or ramp-down periods? Consider filtering those out of the sensor data.
+ We recommend that you avoid using data that contains monotonically increasing values, such as operating hours or mileage.
# Consulting subject matter experts
Lookout for Equipment identifies patterns in the dataset that help to detect critical issues, but it's the responsibility of a technician or subject matter expert (SME) to diagnose the problem and take corrective action, if needed. To ensure that you are getting the right output, we highly recommend that you work with a SME. The SME should help you make sure that you are using the right input data and that your output results are actionable and relevant.