US-20260127556-A1 - AUTOCORRECTIVE FAILURE PREDICTIONS AND ASSISTANCE FOR COMMERCIAL BUILDINGS

US20260127556A1US 20260127556 A1US20260127556 A1US 20260127556A1US-20260127556-A1

Abstract

An example method for building equipment failure prediction and autocorrection, the comprises receiving building management system data associated with building equipment at a site managed by a building management system, analyzing the building management system data or a derivative of the building management system data using a combination of: i) a physics model and ii) a machine-learning model to generate an asset health score (AHS) via the physics model; and a machine-learning score (MLS) via the anomaly detection model; determining an aggregated building equipment failure probability as a function of the AHS and the MLS; based on the aggregated building equipment probability, identifying a building equipment failure of one or more of the building equipment; determining whether the building equipment failure can be automatically remediated; and automatically remediating the building equipment failure.

Inventors

Harsha Mathur
Bhuvaneshwaran K
Praveen Garikapati
Subha Agrawal
Madhav Kamath
Manu Taranath
Navneet KUMAR
Prabhat Ranjan

Assignees

HONEYWELL INTERNATIONAL INC.

Dates

Publication Date: 20260507
Application Date: 20241217
Priority Date: 20241101

Claims (20)

1 . A method for building equipment failure prediction and autocorrection, the method comprising: receiving, by a computing device of a building management system, building management system data associated with building equipment at a site managed by the building management system; analyzing the building management system data or a derivative of the building management system data using a combination of: i) a physics model and ii) an anomaly detection model to generate: an asset health score (AHS) via the physics model; and a machine-learning score (MLS) via the anomaly detection model; determining an aggregated building equipment failure probability as a function of the AHS and the MLS; based on the aggregated building equipment probability, identifying a building equipment failure of one or more of the building equipment; determining whether the building equipment failure can be automatically remediated; and responsive to: determining the building equipment failure can be automatically remediated, initiating an automatic remediation of the building equipment failure; or determining the building equipment failure cannot be automatically remediated, initiating a manual remediation of the building equipment failure.
2 . The method of claim 1 , wherein initiating the automatic remediation comprises initiating a closed-loop control action for remediation of the building equipment with the building equipment failure.
3 . The method of claim 1 , wherein initiating the manual remediation comprises generating a remediation alert, generating an electronic ticket, or both.
4 . The method of claim 1 , wherein the anomaly detection model is a univariate or multivariate anomaly detection model that is configured to detect anomalies in key performance indicators (KPIs) associated with the building equipment.
5 . The method of claim 4 , further comprising querying a plurality of trained supervised machine-learning models that are trained to predict the building equipment failure based on the AHS and the MLS.
6 . The method of claim 1 , wherein the AHS further comprises an AHS for one or more of a heat exchanger, a chiller, a pump, a valve, a fan, a filter.
7 . The method of claim 1 , wherein the aggregated building equipment probability unweighted.
8 . The method of claim 1 , wherein the aggregated building equipment probability is a weighted aggregated building equipment probability that is a function of a weighted average of the AHS and MLS.
9 . The method of claim 8 , wherein the weighted average is at least initially configured as an equally weight average of the physics-based model and the data-drive model, and, wherein the weighted average is adjustable based on a model lifecycle of a plurality of trained supervised learning models that are trained to predict the building equipment failure based on the AHS and the MLS.
10 . The method of claim 8 , wherein the aggregated building equipment failure probability (AGBEIP) is determined in accordance with equation 1 (Eq. 1): AGBEIP=[( W 1 )*(the AHS)]+[( W 2 )*(the MLS)], Eq. 1: wherein “W 1 ” and “W 2 ” are weights attributed to the AHS and MLS, respectively.
11 . The method of claim 1 , further comprising: identifying a plurality of predicted failures; determining a plurality of remediation actions corresponding to the plurality of predicted failures; determining a respective priority of each remediation action of the plurality of remediation actions; and automatically initiating the plurality of remediation actions to remediate the plurality of predicted failures in accordance with the respective priorities of each of the plurality of remediation actions.
12 . The method of claim 11 , wherein the priority of each remediation action of the plurality of remediation actions is based on: a criticality of the building equipment having the predicted failure, an impact of the predicted failure on upstream and/or downstream building equipment, a quantity of hours of operation of the building equipment with the predicted failure, a degree of deviation of the building equipment from a set point of the building equipment, and a historical failure frequency of the building equipment having the predicted failure, or any combination thereof.
13 . The method of claim 1 , further comprising generating a report including information of: a quantity of predicted failures; a type of predicted failures; identifiers of building equipment having a predicted failure; locations of building equipment having a predicted failure; a quantity of remediation actions including a quantity of completed remediation actions, a quantity of pending remediation actions, a quantity of on-going remediation actions, or any combination thereof; a type of remediation actions including type of completed remediation actions that have been completed, a type of pending remediation actions, a type of on-going remediation actions, or any combination thereof, and any combination thereof.
14 . The method of claim 1 , wherein the aggregated building equipment failure probability corresponds to an aggregated building equipment failure probability of an occurrence of a building equipment failure at future time period.
15 . A computing device for building equipment failure prediction and autocorrection, the computing device comprising: a display; a memory; and a processor configured to execute executable non-transitory computer readable instructions stored in the memory to: receive real-world building management system data associated with building equipment at a site managed by the building management system; analyze the real-world data or a derivative of the building management system data using: a physics model to generate an asset health score (AHS); and an anomaly detection model to generate a machine-learning score (MLS); query a plurality of trained supervised machine-learning model with the AHS and the MLS to obtain an aggregated building equipment failure probability; based on the aggregated building equipment failure probability, determine an occurrence of a future building equipment failure; determine whether future building failure can be automatically remediated; and responsive to a determination that the building failure can be automatically remediated, automatically initiating a remediation action to remediate the future building equipment failure.
16 . The computing device of claim 15 , wherein the supervised machine-learning model comprises a deep neural network (DNN).
17 . A non-transitory, computer-readable medium including instructions that when executed by a processor cause the processor to: receive real-world building management system data associated with building equipment at a site managed by the building management system; analyze the real-world building management system data or a derivative of the building management system data using: a physics model to generate an asset health score (AHS); and an anomaly detection model to generate a machine-learning score (MLS); query a plurality of trained supervised machine-learning model with the AHS and the MLS or a weighted average of the AHS and the MLS to obtain an aggregated building equipment failure probability; based on the aggregated building equipment failure probability, determine an occurrence of a future building equipment failure; determine, by querying a large language model (LLM) with at least the determined occurrence of the future building equipment failure, additional information associated with the future building equipment failure, wherein the additional information includes information indicative of whether future building equipment failure can be automatically remediated and optionally includes an indication of remediation to remedy the future building equipment failure; and responsive to a determination that the building failure can be automatically remediated, automatically initiate a remediation action to remediate the future building equipment failure.
18 . The medium of claim 17 , further comprising querying a trained clustering model trained to determine a quantity, a type, or both of a quantity and a type of the plurality of supervised machine-learning models, wherein the trained clustering model is trained at least on a type of potential building equipment failures associated with the building equipment.
19 . The medium of claim 18 , further comprising: determining a building equipment hierarchy including information indicative of upstream and downstream building equipment included in the building equipment hierarchy; and determining the aggregated building equipment failure probability based on the building equipment hierarchy.
20 . The medium of claim 17 , further comprising: displaying, via a display, a recommendation of a remediation action for the building equipment failure; initiating the recommended remediation action for the building equipment failure; and responsive to completion of the remediation action, providing feedback associated with a success of the remediation action to the plurality of trained supervised machine-learning model.

Description

RELATED APPLICATIONS This application claims the benefit of Indian Provisional Patent Application No. 202411083704, filed Nov. 1, 2024, which application is incorporated by reference herein. TECHNICAL FIELD The present disclosure relates to systems, devices, and methods for autocorrective failure predictions and assistance for commercial buildings. BACKGROUND Industrial process control and automation systems are often used to automate large and complex industrial processes. These types of systems routinely include sensors, actuators, and controllers. The controllers are often arranged hierarchically in a control and automation system. For example, lower-level controllers are often used to receive measurements from the sensors and perform process control operations to generate control signals for the actuators. Higher-level controllers are often used to perform higher-level functions, such as planning, scheduling, and optimization operations. Human process operators routinely interact with controllers and other devices in a control and automation system, such as to review warnings, alarms, or other notifications, and adjust the control or initiate performance of other operations (e.g., maintenance operations) to keep the process within desired process limits. If not properly managed, a building equipment failure such as a failure or malfunction of building equipment could escalate into an emergency, crisis, and/or disaster. Moreover, it is essential to monitor the health and functioning of various building equipment, for instance, via building management systems (BMS) systems to ensure optimal performance and achieve energy efficient goals in buildings. Yet, at many locations, building owners may still rely on reactive maintenance (e.g., responsive to the occurrence of future building equipment failure) for fixing future building equipment failure with various building equipment (e.g., HVAC, pumps, heat exchanges, etc.). Relying solely on reactive maintenance may result in extended downtime, increased costs, and/or may negatively impact an experience of tenants and/or occupants of a building (e.g., an HVAC system experiencing a failure may lead to thermal discomfort among building occupants). Additionally, some current dashboards may provide insights into the historical performance and real-time status of the building equipment. However, to forecast when building equipment such as an elevator might require maintenance or when the HVAC system might experience a future issue (e.g., fail), manual analysis of historical data, occupancy schedules, etc. is required. This manual process is time-consuming and often inaccurate, leading to unexpected breakdowns and inefficient asset utilization. As such, some previous approaches may attempt to predict building equipment failures (e.g., prior to an actual occurrence of an actual failure of the building equipment) and/or prior to the failures escalating into serious issues/multiples failures. For instance, some approaches may use as a series of predefined rules (e.g., based on a number of hours of operation of building equipment, etc.) that are configured to identify potential failures. However, such approaches are inflexible and may be inaccurate. Moreover, in such approaches a building equipment failure that is identified can result in a manual remediation by a technician. However, the manual remediation of building equipment failures is time-consuming, costly, and may be ineffective (e.g., the failures may remain unresolved for a longer duration of time and/or a root or contributory cause (e.g., another failure with building equipment that is upstream/downstream) of the failure may remain unresolved (e.g., be undetected and/or unremedied). SUMMARY The present disclosure relates generally to systems, devices, and methods for autocorrective failure predictions and assistance for commercial buildings. The systems, devices, and methods can thereby yield enhanced analytics (e.g., contextualization), prediction, and remediation (e.g., autocorrection) of future building equipment failures such as those related to an industrial process control and automation system. For instance, the systems, devices, and methods herein can employ a combination of a rules-based model, (e.g., to calculate an assert health score (AHS), an anomaly detection model to determine an MLS. An aggregated building equipment failure indicator can be derived from AHS and MLS, as detailed herein. The aggregated building equipment failure indicator can be used to query a plurality of supervised machine-learning models to predict probability of an occurrence of a predicted building equipment failure. In some embodiments, the AHS and the MLS (or the respective probabilities associated with the AHS and the MLS) can be unweighted. Stated differently, the AHS and the MLS (or respective probabilities based on the AHS and the MLS) can be used together as an aggregated building equipment failure indicator to query