Neuromorphic Battery Management System (nBMS)

Abstract

This invention relates to a neuromorphic battery management system (100) and method that monitors and manages operating parameters such as voltage, current and temperature of battery modules used in real time, especially in electric and hybrid vehicles, energy storage systems and energy-critical applications (medical implants, unmanned aerial vehicles, portable artificial intelligence assistants, smart glasses, wearable devices etc.). Within the scope of the invention, the measurement data obtained by the sensor interface circuit (2) and analog / digital converter (3) connected to the battery module (1) are converted into an event-based pulse sequence by the spike encoder unit (4), or in some implementations, the sensor interface circuit (2) is implemented in a neuromorphic / event-based sensor structure, and changes in battery parameters are directly generated as spike sequences and applied to the spike neural network core (5). The outputs obtained from the spike neural network core (5) enable the control of the charge / discharge switches (8) and cell balancing circuits of the battery module (1) through the spike decoder or post-decision processor unit (11) and the gate driver unit (6). The sensor interface circuit (2), analog / digital converter (3), spike encoder unit (4), spike neural network core (5), gate driver unit (6), configuration memory (7), communication interface (9) and power management unit (12) can be implemented integrated on a single neuromorphic battery management integrated circuit (10), thanks to the event-based data processing and sleep / wake-up mechanisms provided by the power management unit (12), low power consumption, high integration and improved battery protection and status prediction functions are achieved, especially in energy-critical applications.

Description

1 TARIFF NEUROMORPHIC BATTERY MANAGEMENT SYSTEM (nBMS) Technical Area This invention has applications in energy storage systems, unmanned aerial vehicles, medical implants, wearable devices, portable AI assistants, smart glasses, military applications and similar energy-critical or portable 5 The invention relates to a battery management system for battery modules used in electronic platforms. It uses neuromorphic computation and spiked neural networks (SNN) in its management, a system-on-a-chip (SoChip) battery with low power consumption It is related to the management system and method. In this description, the neuromorphic battery management system covered by the invention is generally referred to as "neuromorphic battery 10". It will be referred to as a "management system" or "nBMS" for short. The nBMS architecture has different application areas. The product family derivatives are commercially available, for example, “nBMS-Aero” for unmanned aerial vehicles, biomedical and “nBMS-Bio” for medical implant applications and wearable devices, portable artificial intelligence assistants, and Smart glasses may be referred to by names such as “nBMS-Wear”; however, these names do not reflect the technical aspects of the invention. It is not of a limiting nature in terms of its scope. 15 State of the Art Current battery management systems typically measure quantities such as battery cell voltage, current, and temperature. measuring sensors, one or more analog / digital converters connected to these sensors, and processing the measurements. It includes a microcontroller or digital signal processor that controls the charge / discharge switches. This type Systems include cell balancing, overcurrent / voltage / temperature protection, state of charge (SoC), and state of health (SoH) 20 It performs basic BMS functions, such as estimation, using software-based algorithms. In the current state of the technology, machine learning, such as artificial neural networks, is used to predict battery status. Solutions that utilize these approaches exist. However, these solutions are based on von Neumann architecture processors. in the form of algorithms that operate with continuous time steps and have high computational and memory loads. This structure is implemented, especially in energy-critical applications (e.g., applications with limited battery capacity). unmanned aerial vehicles, medical implants placed inside the body, wearable sensor nodes, portable artificial intelligence assistants, smart glasses, military applications, and augmented / mixed reality devices) are real 2 This leads to conflicts between timely response time requirements and very low power consumption targets. It can open. In traditional BMS architectures, sensor data is typically sampled at a fixed frequency, with each sample... Data is retrieved and processed from all cells during this period. Cell parameters are often processed slowly. Despite the changes, high-frequency sampling creates unnecessary load on the processor, which is 5 This leads to a need for a higher clock frequency and consequently higher power consumption. Energy is critical. In these applications, this unnecessary power consumption has a direct impact on the overall operating time and reliability of the system. It has a negative impact. Furthermore, current state solutions to the technology address issues such as cellular senescence, capacity loss, and temperature-dependent behavior. This involves modeling nonlinear and time-dependent effects with a limited number of parameters, these 10 This situation leads to errors between the model and actual behavior in field conditions, and delays in early fault detection. This can cause problems. Neuromorphic computation and spiny neural networks (SNNs), event-based data processing, low power It is known as a new generation computing paradigm that provides low power consumption and high parallelism. In the current state of the technique, tight integration of the neuromorphic nucleus with the battery management system is 15. It is built on SoChip and operates end-to-end event-based, from the sensor interface to the door drivers. Integrated, bespoke architectures are limited, especially for energy-critical applications. Purpose of the Invention The main purposes of this invention are listed below: • Battery management functions (SoC / SoH estimation, cell balancing, error and anomaly detection, 20 (security protections, etc.) a low-power, event-based, and real-time neuromorphic architecture to ensure it is carried out. • Battery sensor data is converted to event-triggered spike signals instead of continuous sampling. By transforming and processing it, we can reduce bus traffic and computational load; thus especially medical implants, unmanned aerial vehicles, wearable devices, portable artificial intelligence 25 Battery management in energy-critical systems such as assistants, smart glasses, military applications and similar devices. To minimize the power consumption of the unit. 3 • Complex, nonlinear, and age-related behaviors of battery cells, spinous nerves to implement this on an SoChip that can learn with its networks and thus achieve more accurate SoC / SoH To predict and detect faults early. • BMS functions, measurement interfaces, spike encoders, neuromorphic kernel, configuration to integrate the memory and door driver blocks into a single system-on-a-chip architecture; thus 5 To reduce the card area and the number of external components. • Separate control of battery charge / discharge switches via neuromorphic core outputs by providing cell-based precision balancing and safety mechanisms. • Especially for systems equipped with event-based sensors, event-based system integrity Managing battery functions without disrupting them. 10 • All these features are used in medical implants, unmanned aerial vehicles, wearable devices, and portable artificial intelligence. Intelligence assistants, smart glasses, military applications and similar devices with limited energy resources To create a battery management solution suitable for use across platforms. The neuromorphic battery management system and method developed for this purpose is a classic approach in the current state of the technique. Compared to digital BMS architectures, it offers significantly lower power consumption and optimally scheduled calculations, resulting in high performance. It offers a structure that governs with integrity. Brief Description of the Figures Figure 1: General view of the neuromorphic battery management system (100) in block diagram form. Figure 1A: Block diagram of an alternative nBMS architecture (100) containing neuromorphic or event-based sensors. appearance. 20 Figure 2: Detailed schematic of the internal block architecture of the neuromorphic battery management SoChip (10). Figure 3: Example topology of the spike neural network core (5) and spike encoder / decoder units (4, 11). diagram showing. Figure 4: Process steps showing the flowchart of the neuromorphic battery management method (101–110). Figure 5: In an energy-critical application (e.g., medical implant, drone, wearable device, 25 Neuromorphic battery management used in portable artificial intelligence assistants or smart glasses Diagram showing the layout of the system (100). 4 Explanation of References in Figures 1: Battery module 2: Sensor interface circuit (may include analog or neuromorphic / event-based sensor architecture) 3: Analog-to-Digital Converter (ADC) 4: Spike encoder unit 5 : Spiky neural network (SNN) kernel 6: Door driver unit 7: Configuration and parameter memory 8: Charge / discharge switches (switching elements) 9: Communication interface (e.g., CAN, LIN, UART, I2C, etc.) 10 Neuromorphic Battery Management Integrated Circuit (SoChip) 11: Thorn remover or post-decision processing unit 12: Power management and clock generation unit 13: Auxiliary microcontroller core (optional) 14: Energy-critical application unit (e.g., medical implant, unmanned aerial vehicle, wearable device, portable 15 artificial intelligence assistant, smart glasses or augmented / mixed reality glasses) 100: General overview of neuromorphic battery management systems. 101–110: Blocks showing the method steps Detailed Description of the Invention This section provides a detailed description of a prototype implementation of the invention. The 20 given here... The example implementation does not limit the scope of the invention; the invention can be defined by different component choices and It can also be implemented with parameters. The neuromorphic battery management system described in this invention is generally referred to as "nBMS" in practice. It can be customized. The nBMS architecture can be customized according to the application scenario, for example, nBMS-Aero for unmanned aerial vehicles. nBMS-Bio for biomedical and medical implant systems and wearable devices, portable artificial intelligence 25 This can be implemented in the form of product family variants such as nBMS-Wear for assistants and smart glasses. The descriptions below outline the core technical features common to all of these variants. Neuromorphic battery management system (100) sensor interface circuit connected to at least one battery module (1) (2), the analog / digital converter (3) connected with this circuit, digitizes the measurement data. The spike encoder unit (4) which converts into sequences, and the spike neural network connected to the spike encoder unit (4) 5 its core (5), gate located between the spiny neural network core (5) and the battery charge / discharge switches (8) It includes the driver unit (6) and the configuration memory (7) which stores the operating parameters of the system. The technical connections between these components and their positions within the system are shown in a block diagram in Figure 1. It is shown in the form of a significant portion of these components, a single neuromorphic battery management integrated circuit (10) It was implemented in an integrated manner. 10 The battery module (1) contains numerous electrochemical cells connected in series and / or parallel. Each cell or Voltage, current and temperature measurements for the cell group are taken via the sensor interface circuit (2). Sensor The interface circuit (2) may include, for example, multiplexer, preamplifier or galvanic isolation units. Energy In critical applications (14), the battery module (1) typically has limited capacity, and the total energy of the system A portion of its budget is sensitive to the consumption of the battery management unit. 15 Analog signals from the sensor interface circuit (2) are digitized by the analog / digital converter (3). ADC (3) can take continuous time measurements as well as depending on certain threshold changes in battery status. It can also perform triggered event-based sampling. For example, when the cell voltage exceeds a certain value... If the level drops or changes to a certain level, a new sampling can be triggered. This allows the battery to... When the parameters do not change for a long time, ADC (3) operates at a low sampling rate or sleep mode 20 It reduces power consumption. The spike encoder unit (4) digitizes the measurement data, which is the output of the ADC (3), into the spike neural network core. Converts into a sequence of event pulses (spikes) located in time, in a suitable format for (5). Spikes coding, for example, rate coding, time-based threshold coding, or event-based threshold overshoot coding This can be implemented in this way. In this way, when battery parameters change infrequently, the system receives a low number of spikes. It produces and the neuromorphic nucleus (5) remains largely in low power mode; when the parameters change rapidly At certain moments, a more intense spike flow occurs, and the system responds quickly to dynamic events. Thus, especially Average power consumption is significantly reduced in energy-critical applications. In some alternative implementations, as shown in Figure 1A, the sensor interface circuit (2) is in the classical sense. Instead of an analog front end, changes in at least one parameter of the battery module (1) can be directly triggered or 30 6 in a neuromorphic or event-based sensor structure that produces pulse sequences This can be implemented. In this case, the sensor interface circuit (2) detects the battery voltage, current or temperature. changes that generate spike signals when certain thresholds are exceeded or the rate of change surpasses certain limits. It may include a sensor matrix or comparator network. In this neuromorphic sensor configuration, analog / digital Sampling and encoding performed in the converter (3) and / or spike encoder unit (4) 5 Some or all of its functions can be shifted into the sensor interface circuit (2); in this case the sensor The spike arrays generated by the interface circuit (2) are directly applied to the spike neural network core (5) and The ADC (3) and the spike encoder unit (4) can be disabled or simplified. The spiny neural network core (5) is a hardware containing a network of numerous interconnected spiny neurons and synapses. It is an accelerator-driven neuromorphic processing unit. This core receives spike sequences from sensors, battery 10 cell charge status (SoC), health status (SoH), internal resistance change, temperature profile, and potential failures. The neurons located in the internal structure of the spiny neural network nucleus (5) produce indicator signals about their condition. It can form a layered or iterative network structure. This neuron topology handles the processing of spindle inputs and It converts the effect of synaptic weights into a functional neural flow as shown in Figure 3. Spiky neural network The core (5) retrieves the synaptic weights obtained during training from the configuration memory (7) and study 15 During this process, it makes inferences based on these weights. In some implementations, the kernel (5) performs online adjustment. Or it can be structured to allow for limited on-site learning. Configuration and parameter memory (7) synaptic weights of spiny neural network nuclei (5), neuron its parameters, the threshold and scale parameters of the spike encoder unit (4), the battery module (1) It stores configuration information and safety limit values. This memory can be, for example, an embedded flash memory or 20 It could be EEPROM. Output indicators generated by the spike neural network core (5) are used by the spike resolver or post-decision processor. The unit (11) processes the SoC / SoH estimates into numerical values, cell-based. It generates balancing commands or provides protection against situations such as overcurrent, overvoltage, and overtemperature. It makes decisions. 25 The door driver unit (6) is connected to the battery module (1) according to the commands it receives from the decision post-processor unit (11). It drives the charge / discharge switches (8). The switches (8) can be MOSFET or IGBT, for example, and both battery current It can be used to both open and close the path, as well as to control cell-based balancing circuits. 7 The communication interface (9) is between the neuromorphic battery management integrated circuit (10) and external control units. It enables data communication. Battery status information, warnings, and errors can be transmitted through this interface, System parameters and SNN weights can also be updated. Power management and clock generation unit (12) blocks inside the neuromorphic battery management integrated circuit (10). It provides the necessary supply voltages and clock signals. Event-based structure of the neuromorphic kernel (5) 5 thanks to the power management unit (12) neuromorphic nucleus and spike encoding unit (4), battery It can enter low power or sleep mode when the parameters do not change, and only activate these blocks when an event occurs. It can wake you up. In some implementations, the neuromorphic battery management integrated circuit (10) includes diagnostic and communication functions. 10 A helper microcontroller core (13) to manage protocols and configuration operations It can be found. The basic process steps of the neuromorphic battery management method are shown in a flowchart in Figure 4. The method is shown. The voltage, current and temperature data of the battery module (1) are obtained from the sensor interface circuit. (2) and its acquisition with ADC (3) (101), these data are converted into spike arrays by the spike encoder unit (4) transformation (102), processing of spike sequences in the spike neural network core (5) (103), core (5) 15 (104) and evaluation of the outputs by the spike remover or post-decision processing unit (11) and gate It includes the control of the charge / discharge switches (8) via the driver unit (6) (105). In the implementation of an alternative method, at least one parameter of the battery module (1) is neuromorphic or event- is produced directly in the form of spikes or pulse sequences by a sensor interface (2) based on these spikes The sequences are passed through the analog / digital converter (3) and the spike encoder unit (4) without the spike neural network 20 It is applied to the core (5). In this case, the method involves sampling and spike coding steps at the sensor interface. Thanks to the implementation within the circuit (2), the event on the sensor side is also constant while the battery parameters are constant. By reducing production, power consumption can be further decreased. In some implementations for energy-critical applications, the method shows a significant improvement in battery parameters. Unless there is a change, the spike encoding and neuromorphic processing steps occur at a low frequency or during sleep 25 the system should be kept in this mode, with spike signal generation and extraction only occurring when certain thresholds are exceeded. This allows the average power consumption of the neuromorphic battery management integrated circuit (10) to be further reduced. 8 How the invention can be applied to industry. The invention concerns a neuromorphic battery management system and method for electric vehicles and stationary energy storage. systems, industrial power systems, portable electronic devices, and especially in energy-constrained applications. It can be used as a battery management unit. Within the scope of energy-critical applications; medical implants (e.g., pacemakers, neurostimulators, in-body implants) sensor nodes, etc.), in small form factor unmanned aerial vehicles, wearable health monitoring systems, in portable AI assistants, smart glasses, augmented or mixed reality glasses, found in wireless sensor networks, military applications, and other closed systems that cannot be accessed remotely. It can be used in the management of battery packs. In these applications, event-based and neuromorphic architecture is used. Thanks to its low power consumption design, the battery management unit accounts for a much smaller percentage (10%) of the system's total energy consumption. It has a small share and the service life of the battery-powered device is extended. In applications equipped with event-based sensors or designed directly as event-based It can be used directly and easily for event-based battery management in systems without compromising system integrity. It is an adaptable battery management system. Product family derivatives adapted to different application areas of nBMS architecture; unmanned aerial vehicles and flying 15 nBMS-Aero for platforms, nBMS-Bio for biomedical and medical implant systems, and wearable for devices, portable AI assistants, smart glasses and similar portable consumer electronics It may be offered under trade names such as nBMS-Wear. Such designations are similar to neuromorphic battery management. refers to versions of the kernel adapted to different mechanical, electrical, or software interfaces. It is. 20 Neuromorphic battery management integrated circuit (10) integrated using standard semiconductor manufacturing processes. It can be manufactured in circuit form and soldered onto a printed circuit board, becoming a component of the battery management board. It can be used as. The invention utilizes currently used analog sensors, MOSFET / IGBT switches, and Since it can be implemented in a way that is compatible with communication protocols, it is relatively superior to existing BMS designs. It can be integrated with minimal hardware changes. 25

Claims

9 REQUESTS 1. The invention describes a neuromorphic system that monitors the voltage, current and temperature values ​​of at least one battery module (1). The battery management system (100) has the feature of receiving analog signals from sensors connected to the battery module (1). a sensor interface circuit (2) that processes the signals, analog signals from the sensor interface circuit (2) At least one analog / digital converter (3) digitizing the digitized measurement data event-5 at least one spike encoder unit (4) which converts to a series of pulses based on the spike encoder unit (4) By processing the incoming shocks, the charge status, health status and safety of the battery module (1) A spiny neural network core (5) that produces indicators is connected to the output of the spiny neural network core (5). as a battery module (1) that generates control signals to drive the charge / discharge switches (8) Synaptic weights of gate driver unit (6) and spiny neural network core (5) and system operation 10 It contains a configuration memory (7) that stores its parameters.

2. The invention describes a neuromorphic system that monitors the voltage, current and temperature values ​​of at least one battery module (1). The battery management system (100) has the feature that at least one parameter of the battery module (1) neuromorphic or changes that produce a series of direct pulses (spikes) instead of an analog signal. A sensor interface circuit (2) containing an event-based sensor structure, by sensor interface circuit (2) 15 The generated spike or pulse sequences cannot be converted using any analog / digital conversion or separate process. By processing directly without requiring a coding phase, the charge status of the battery module (1) is processed, a spiny neural network core (5) that produces health status and safety indicators, spiny neural network depending on the output of the core (5) to drive the charge / discharge switches (8) of the battery module (1). a gate driver unit (6) that generates control signals and the synaptic 20 of the spiny neural network core (5) It contains a configuration memory (7) which stores the weights and system operating parameters.

3. Neuromorphic battery management system (100) according to claim 1 or 2, and its feature is; sensor interface separate voltage and temperature for each cell or group of cells in the (2) battery module (1) of the circuit It includes measurement channels.

4. Neuromorphic battery management system (100) according to claim 1, 2 or 3, and its feature is; analog / digital 25 sampling frequency depending on the rate of change in the battery parameters of the converter (3) dynamically adjustable or event-based sampling when certain threshold values ​​are exceeded It is a triggering structure.

5. Neuromorphic battery management system (100) according to any of claims 1–4, and its feature is; thorn The encoder unit (4) digitizes the measurement data into at least one threshold-based encoding scheme. conversion of pulses into a sequence positioned in the time domain and battery parameters When not changed, it reduces spike production, thereby lowering the system's power consumption.

6. Neuromorphic battery management system (100) according to any of claims 1–5, and its feature is; spiny The neural network core (5) contains a network topology consisting of numerous spiny neurons and synapses and The charge status and health status of the battery module (1) are stored in the configuration memory (7). It is a real-time prediction using synaptic weights.

7. Neuromorphic battery management system (100) according to any of claims 1–6, and its feature is; gate the safety indicators generated by the spiny neural network core (5) of the driver unit (6) by switching the charge / discharge switches (8) on and off, overcurrent, overvoltage and overtemperature In these situations, it protects the battery module (1) and controls the cell-based balancing circuits. 10 8. Neuromorphic battery management system (100) according to any of claims 1–7, and its feature is; sensor interface circuit (2), analog / digital converter (3), spike encoder unit (4), spike neural network core (5), door driver unit (6), configuration memory (7) and an optional communication interface (9) and the power management unit (12) on a single neuromorphic battery management integrated circuit (10) It is carried out in an integrated manner. 15 9. Neuromorphic battery management system (100) according to any of claims 1–8, and its feature is power. the management unit (12) spin encoder unit (4) and spin neural network core (5) battery When there are no changes in its parameters, it enters low power or sleep mode and only uses sensors. Depending on the threshold overshoot events from the interface circuit (2) and the analog / digital converter (3), this By enabling the units, the average power of the neuromorphic battery management integrated circuit (10) is 20 The goal is to maintain energy consumption at a level suitable for energy-critical applications.

10. Neuromorphic battery management system (100) according to any of claims 1–9, and its feature is; The neuromorphic battery management integrated circuit (10) has an energy-critical application unit (14), for example a medical implant, unmanned aerial vehicle, wearable device, portable artificial intelligence assistant or smart It is used together with the battery module (1) of the glasses. 25 11. The invention describes a neuromorphic architecture for the voltage, current and temperature values ​​of at least one battery module (1). a neuromorphic battery management system that enables battery management by processing it The method is characterized by the fact that the voltage, current and temperature values ​​of the battery module (1) are displayed on the sensor interface. (101), receiving through circuit (2) and digitizing with analog / digital converter (3), The digitized measurement data is converted into an event-based pulse sequence by the spike encoder unit (4) 30 (102) transformation of the sequence of pulses from the spike encoder unit (4) into a spike neural network 11 By applying it to the core (5), the charge status, health status and safety of the battery module (1) Calculation of indicators (103), spiny neural network core (5) outputs or spiny solver or Evaluation by the post-decision processing unit (11) and check for charge / discharge switches (8). producing decisions (104), door according to control decisions from post-decision processing unit (11) Switching of the charge / discharge switches (8) of the battery module (1) via the driver unit (6) and 5 If necessary, the driving of the cell balancing circuits consists of (105) steps.

12. According to claim 11, the neuromorphic battery management method is characterized by its spike encoding unit (4) It adjusts the spike production density depending on the rate of change in battery parameters and By reducing spike production when battery parameters remain unchanged, the power of the spike neural network core (5) is reduced. It is to reduce consumption.10 13. Neuromorphic battery management method according to claim 11 or 12, characterized by its spiny neural network. offline training based on the measurement data collected in the field conditions of the core (5) retrieve the updated synaptic weights from the configuration memory (7) and the updated weights It includes the step of uploading it to the system via the communication interface (9).

14. A neuromorphic battery management method according to any of claims 11–13, characterized by; battery15 Spike coding (102) and during the period when the predetermined thresholds in its parameters are not exceeded suppression of spiny neural network processing (103) steps and neuromorphic battery management during this time Keeping the integrated circuit (10) in low power or sleep mode by the power management unit (12), These steps are only activated when threshold exceedance events are detected. The goal is to reduce average power consumption to a level suitable for energy-critical applications.20 15. A neuromorphic battery management method according to any of claims 11–14, characterized by its spike-like properties. Anomaly and failure patterns occurring in the battery module (1) of the neural network core (5) event- It detects based on and safely charges / discharges the switches (8) via the door driver unit (6). This involves taking the step of electrically isolating the battery module (1) by putting it into mode.

16. A neuromorphic battery management method according to any of claims 11–13, characterized by; battery25 Neuromorphic or event-based sensor interface circuit (2) for at least one parameter of module (1) implemented in a sensor structure, the spikes correspond to changes in the parameters in question. or directly generate pulse sequences and these spike sequences are sent to the spiny neural network core (5) It includes the implementation step.

Images