Apparatuses, methods, and computer program products for energy-centric predictive maintenance scheduling

1 . A method, comprising:

at a device with one or more processors and one or more memories:

inputting historical time-varying sensor state values from one or more sensors associated with an asset along with historical time-varying asset-independent data and categorical features associated with the asset into a data model to train the data model, wherein the historical time-varying sensor state values and the historical time-varying asset-independent data are time-matched to train the data model;

training the data model using the historical time-varying sensor state values, the historical time-varying asset-independent data, and the categorical features to derive one or more trained model weights;

retraining the trained data model based on the one or more trained model weights using historical data associated with at least one asset similar to the asset, wherein the trained data model is deployed on a cloud, and wherein at least one operation of the asset and the one or more sensors is controlled by at least one respective controller associated with the asset;

identifying at least one of an upstream asset and a downstream asset associated with the asset, wherein the asset is tagged with one or more data values that indicate association between the asset and the at least one of the upstream asset and the downstream asset, and wherein a failure associated with the asset affects the at least one of the upstream asset and the downstream asset;

inputting expected future time-varying asset-independent data over a time frame and the one or more data values into the trained data model;

generating from the trained data model predicted sensor state values associated with the asset over the time frame;

generating from the trained data model predicted energy usage by the asset over the time frame;

determining optimum energy usage by the asset over the time frame;

calculating energy wastage over the time frame based on a difference between the predicted energy usage and the optimum energy usage of the asset over the time frame;

calculating, using the predicted sensor state values, one or more asset performance metrics corresponding to one or more preventive maintenance tasks;

generating one or more recommended service tasks over the time frame based at least in part on the one or more calculated performance metrics and at least in part on the calculated energy wastage over the time frame;

automatically controlling, based on the generated recommended service tasks, operation of the asset via the at least one respective controller associated with the asset by adjusting at least one of: a mode of operation, a chilled water supply set point, activation or manipulation of an actuator of the asset to reduce predicted energy usage until the recommended service task is performed; and

responsive to the controlling operation of the asset and based at least in part on the generated recommended service tasks, automatically rendering a user interface to display the one or more recommended service tasks for the asset along with an associated timing for performing the recommended service tasks and a bar graph showing remaining days until each service task of recommended service tasks is due.

2 . The method of claim 1 , wherein the optimum energy usage by the asset over the time frame is determined based on historical energy usage by the asset.

3 . The method of claim 1 , wherein the optimum energy usage by the asset over the time frame is determined based on industry-standard data corresponding to a category of asset to which the asset belongs.

4 . The method of claim 1 , wherein the recommended service tasks are based at least in part on a date when the calculated energy wastage cumulatively equals a cost of the one or more recommended service tasks.

5 . The method of claim 1 , wherein the data model comprises a temporal fusion transformer deep learning model.

6 . The method of claim 1 , wherein the future time-varying asset-independent data comprises at least forecasted weather and occupancy data.

7 . An apparatus comprising at least one processor and at least one non-transitory memory comprising program code, wherein the at least one non-transitory memory and the program code are configured to, with the at least one processor, cause the apparatus to at least:

input historical time-varying sensor state values from one or more sensors associated with an asset along with historical time-varying asset-independent data and categorical features associated with the asset into a data model to train the data model, wherein the historical time-varying sensor state values and the historical time-varying asset-independent data are time-matched to train the data model;

train the data model using the historical time-varying sensor state values, the historical time-varying asset-independent data, and the categorical features to derive one or more trained model weights;

retrain the trained data model based on the one or more trained model weights using historical data associated with at least one asset similar to the asset, wherein the trained data model is deployed on a cloud, and wherein at least one operation of the asset and the one or more sensors is controlled by at least one respective controller associated with the asset;

identify at least one of an upstream asset and a downstream asset associated with the asset, wherein the asset is tagged with one or more data values that indicate association between the asset and the at least one of the upstream asset and the downstream asset, and wherein a failure associated with the asset affects the at least one of the upstream asset and the downstream asset;

input expected future time-varying asset-independent data over a time frame and the one or more data values into the trained data model;

generate from the trained data model predicted sensor state values associated with the asset over the time frame;

generate from the trained data model predicted energy usage by the asset over the time frame;

determine optimum energy usage by the asset over the time frame;

calculate energy wastage over the time frame based on a difference between the predicted energy usage and the optimum energy usage of the asset over the time frame;

calculate, using the predicted sensor state values, one or more asset performance metrics corresponding to one or more preventive maintenance tasks;

generate one or more recommended service tasks over the time frame based at least in part on the one or more calculated performance metrics and at least in part on the calculated energy wastage over the time frame;

automatically control, based on the generated recommended service tasks, operation of the asset via the at least one respective controller associated with the asset by adjusting at least one of: a mode of operation, a chilled water supply set point, activation or manipulation of an actuator of the asset to reduce predicted energy usage until the recommended service task is performed; and

responsive to the controlling operation of the asset and based at least in part on the generated recommended service tasks, automatically render a user interface to display the one or more recommended service tasks for the asset along with an associated timing for performing the recommended service tasks and a bar graph showing remaining days until each service task of recommended service tasks is due.

8 . The apparatus of claim 7 , wherein the optimum energy usage by the asset over the time frame is determined based on historical energy usage by the asset.

9 . The apparatus of claim 7 , wherein the optimum energy usage by the asset over the time frame is determined based on industry-standard data corresponding to a category of asset to which the asset belongs.

10 . The apparatus of claim 7 , wherein the recommended service tasks are based at least in part on a date when the calculated energy wastage cumulatively equals a cost of the one or more recommended service task.

11 . The apparatus of claim 7 , wherein the data model comprises a temporal fusion transformer deep learning model.

12 . The apparatus of claim 7 , wherein the future time-varying asset-independent data comprises at least forecasted weather and occupancy data.

13 . A computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising an executable portion configured to:

input historical time-varying sensor state values from one or more sensors associated with an asset along with historical time-varying asset-independent data and categorical features associated with the asset into a data model to train the data model, wherein the historical time-varying sensor state values and the historical time-varying asset-independent data are time-matched to train the data model;

train the data model using the historical time-varying sensor state values, the historical time-varying asset-independent data, and the categorical features to derive one or more trained model weights;

retrain the trained data model based on the one or more trained model weights using historical data associated with at least one asset similar to the asset, wherein the trained data model is deployed on a cloud, and wherein at least one operation of the asset and the one or more sensors is controlled by at least one respective controller associated with the asset;

identify at least one of an upstream asset and a downstream asset associated with the asset, wherein the asset is tagged with one or more data values that indicate association between the asset and the at least one of the upstream asset and the downstream asset, and wherein a failure associated with the asset affects the at least one of the upstream asset and the downstream asset;

input expected future time-varying asset-independent data over a time frame and the one or more data values into the trained data model;

generate from the trained data model predicted sensor state values associated with the asset over the time frame;

generate from the trained data model predicted energy usage by the asset over the time frame;

determine optimum energy usage by the asset over the time frame;

calculate energy wastage over the time frame based on a difference between the predicted energy usage and the optimum energy usage of the asset over the time frame;

calculate, using the predicted sensor state values, one or more asset performance metrics corresponding to one or more preventive maintenance tasks;

generate one or more recommended service tasks over the time frame based at least in part on the one or more calculated performance metrics and at least in part on the calculated energy wastage over the time frame;

automatically control, based on the generated recommended service tasks, operation of the asset via the at least one respective controller associated with the asset by adjusting at least one of a mode of operation, a chilled water supply set point, activation or manipulation of an actuator of the asset to reduce predicted energy usage until the recommended service task is performed; and

responsive to the controlling operation of the asset and based at least in part on the generated recommended service tasks, automatically render a user interface to display the one or more recommended service tasks for the asset along with an associated timing for performing the recommended service tasks and a bar graph showing remaining days until each service task of recommended service tasks is due.

14 . The computer program product of claim 13 , wherein the optimum energy usage by the asset over the time frame is determined based on historical energy usage by the asset.

15 . The computer program product of claim 13 , wherein the optimum energy usage by the asset over the time frame is determined based on industry-standard data corresponding to a category of asset to which the asset belongs.

16 . The computer program product of claim 13 , wherein the recommended service tasks are based at least in part on a date when the calculated energy wastage cumulatively equals a cost of the one or more recommended service task.

17 . The computer program product of claim 13 , wherein the data model comprises a temporal fusion transformer deep learning model.