Deep-learning-based fault detection in building automation systems

1. A method for fault detection of a building automation system using deep learning comprising the steps of:

maintaining a system manager application in a data processing system of a management system configured to perform building automation system functions and to provide a graphical user interface; and

running the system manager application to:

receive point data for a hardware being analyzed, wherein the received point data is contaminated data;

train a deep learning model for the hardware being analyzed based on the received point data, without supplemental information, by applying a loss function;

generate predicted data based on the deep learning model;

normalize some or all of the predicted data or the received point data;

analyze the predicted data and the received point data by applying cumulative sum control chart (CUSUM) sequential analysis for summing, weighting, and change detection;

identify a fault in the hardware being analyzed according to the received point data and the predicted data; and

produce a fault report according to the identified fault.

2. The method of claim 1 , wherein the received data includes at least required points for the hardware being analyzed.

3. The method of claim 1 , wherein training the deep learning model by applying the loss function includes applying a Huber loss function.

4. The method of claim 1 , wherein training the deep learning model includes applying dropout techniques for regularization.

5. The method of claim 1 , wherein identifying the fault includes comparing the received point data for a first period of time with the predicted data for a corresponding second period of time.

6. The method of claim 1 , wherein identifying the fault includes identifying when a normalized deviation between received point data and the predicted data is greater than a predetermined threshold.

7. The method of claim 1 , wherein the fault report is a graphic user interface illustrating the received point data as compared to the predicted data.

8. A non-transitory computer-readable medium encoded with executable instructions that is configured to run in a data processing system of a management system, configured to perform building automation system functions, and configured to provide a graphical user interface, wherein the building automation system functions include:

receiving point data for a hardware being analyzed, wherein the received point data is contaminated data;

training a deep learning model for the hardware being analyzed based on the received point data, without supplemental information, by applying a loss function;

generating predicted data based on the deep learning model;

normalizing some or all of the predicted data or the received data;

analyzing the predicted data and the received point data by applying cumulative sum control chart (CUSUM) sequential analysis for summing, weighting, and change detection;

identifying a fault in the received point data with respect to the predicted data; and

producing a fault report according to the identified fault.

9. The non-transitory computer-readable medium of claim 8 , wherein the received data includes at least required points for the hardware being analyzed.

10. The non-transitory computer-readable medium of claim 8 , wherein training the deep learning model by applying the loss function includes applying a Huber loss function.

11. The non-transitory computer-readable medium of claim 8 , wherein training the deep learning model includes applying dropout techniques for regularization.

12. The non-transitory computer-readable medium of claim 8 , wherein identifying the fault includes comparing the received point data for a first period of time with the predicted data for a corresponding second period of time.

13. The non-transitory computer-readable medium of claim 8 , wherein identifying the fault includes identifying when a normalized deviation between received point data and the predicted data is greater than a predetermined threshold.

14. The non-transitory computer-readable medium of claim 8 , wherein the fault report is a graphic user interface illustrating the received point data as compared to the predicted data.

15. A building automation system comprising a data processing system, and a plurality of devices, sensors, and actuators, wherein the data processing system includes a graphical user interface and executes a system manager application to perform building automation system functions and to:

receive point data for a hardware being analyzed, wherein the received point data is contaminated data;

train a deep learning model for the hardware being analyzed based on the received point data, without supplemental information, by applying a loss function;

generate predicted data based on the deep learning model;

normalize some or all of the predicted data or the received point data;

analyze the predicted data and the received point data by applying cumulative sum control chart (CUSUM) sequential analysis for summing, weighting, and change detection;

identify a fault in the hardware being analyzed according to the received point data and the predicted data; and

produce a fault report according to the identified fault.

16. The building automation system of claim 15 , wherein training the deep learning model by applying the loss function includes applying a Huber loss function and dropout techniques for regularization.