Brain enlightenment circuit with context dependent memory and state dependent memory and application

Abstract

The invention relates to the field of bionic circuits, and discloses a brain enlightenment circuit with context-dependent memory and state-dependent memory, comprising: a sensory memory module for receiving and encoding information stimulation to obtain information encoding information; the short-time memory module is used for receiving the information coding information output by the sense memory module and determining whether to repeat the information coding information and store the information coding information into long-time memory; the state module is used for receiving the psychological state information and encoding the psychological state information to obtain state encoding information; the situation module is used for receiving the situation information and encoding the situation information to obtain situation encoding information; and the long-time memory module is used for receiving the information coding information repeated by the short-time memory module. Through an adaptive damage memory and extraction mechanism, false alarm and missing alarm can be effectively reduced, the accuracy of industrial robot damage assessment is improved, and the method can meet the requirement of multivariable damage monitoring in a complex production environment.

Description

Technical Field

[0001] The present invention relates to the technical field of bionic circuits, and specifically to a brain-inspired circuit with context-dependent memory and state-dependent memory and its applications. Background Art

[0002] In recent years, inspired by the working mechanism of the brain, brain-inspired intelligence has become an important research direction in artificial intelligence. However, the current von Neumann architecture adopts a mode of separating storage and computing, where data is frequently exchanged between the storage unit and the computing unit, resulting in limited computing efficiency and restricting the parallel processing ability and operating speed of neural networks. In contrast, the brain can store and process a large amount of information simultaneously in a low-power form. Therefore, constructing a neural network with in-memory computing is an important direction to improve computing efficiency.

[0003] As a non-linear resistor with memory ability, the memristor can perform both computing and storage simultaneously. With advantages such as low power consumption and high integration, it is widely used in fields such as associative memory and intelligent robots, and has become a key technology for simulating brain information processing.

[0004] The biological memory system follows a three-level processing model, including sensory memory, short-term memory, and long-term memory. Moreover, the context and mental state during information storage are also encoded, making the information easier to be retrieved in a specific environment or state, which are respectively called context-dependent memory and state-dependent memory. However, existing brain-inspired storage research mainly focuses on information storage, with less research on information retrieval, and cannot effectively simulate the dynamic changes affected by various factors during the information retrieval process. To address this problem, the present invention proposes a brain-inspired circuit that comprehensively considers the degree of information memory, the time interval between memory and retrieval, state factors, and context factors, constructs a complete information encoding, storage, and retrieval mechanism, and extends the application to the damage detection of industrial robots based on this. In industrial scenarios such as automobile manufacturing, robots operate in different environments for a long time and are easily damaged by factors such as high temperature, dust, and humidity. Existing detection methods mostly rely on real-time sensing data, cannot record the damage change trend, and are also difficult to conduct damage analysis in combination with environmental factors. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a brain-inspired circuit with context-dependent memory and state-dependent memory and its applications, which solves the problem that the existing storage and computing architecture is difficult to simultaneously meet the requirements of low power consumption, efficient storage, and intelligent retrieval. The present invention realizes the dynamic optimization of information storage and retrieval by constructing a brain-inspired circuit with context-dependent memory and state-dependent memory, and extends it to the damage detection of industrial robots, improving the intelligence and accuracy of damage assessment.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A brain-inspired circuit with context-dependent memory and state-dependent memory, comprising:

[0007] A sensory memory module for receiving information stimuli and encoding them to obtain information encoding information;

[0008] A short-term memory module for receiving the information encoding information output by the sensory memory module and deciding whether to repeat and store the information encoding information in long-term memory;

[0009] A state module for receiving psychological state information and encoding it to obtain state encoding information;

[0010] A context module for receiving context information and encoding it to obtain context encoding information;

[0011] A long-term memory module for receiving the information encoding information repeated by the short-term memory module and storing the state encoding information of the state module and the context encoding information of the context module in long-term memory;

[0012] An information retrieval module for receiving an information retrieval signal and extracting corresponding stored information from the long-term memory module according to the clue signals provided by the context and / or state, and transmitting the stored information to the short-term memory module.

[0013] Preferably, the output of the short-term memory module is regulated by an attention signal, and this attention signal controls the information learning rate and forgetting rate.

[0014] Preferably, the long-term memory module includes an information long-term memory module, a context long-term memory module, and a state long-term memory module. The information long-term memory module is used to store information encoding information, the context long-term memory module is used to store context encoding information, and the state long-term memory module is used to store state encoding information. And the stored contents of the information long-term memory module, the context long-term memory module, and the state long-term memory module cooperate with each other during information retrieval to improve the success rate of information extraction.

[0015] Preferably, the context module and the state module match with the stored context encoding information or state encoding information by analyzing the current environment or psychological state to output a clue signal, which is used to accelerate information extraction and dynamically adjust the success rate of information retrieval according to the matching degree of the context or state.

[0016] Preferably, the information retrieval module determines whether to accelerate information recall by receiving the information retrieval signal and the clue signal. If the context or state matches, the clue signal is enhanced.

[0017] Preferably, the context module analyzes the information in the current environment and matches it with the stored context coding information, thereby providing a clue signal for information extraction.

[0018] Preferably, the state module receives the signal of the current mental state, compares it with the stored state coding information, and provides a clue signal during information retrieval to accelerate the recall process.

[0019] Preferably, the information includes temperature, humidity, time, and location.

[0020] Preferably, the long-term memory module includes a calculation and feedback module. The calculation and feedback module is used to calculate the matching degree between the current context coding information and state coding information and the stored data through a multiply-accumulate operation circuit, adjust the weight value of the memristor matrix according to the matching result, and provide a feedback signal to the information retrieval module through a digital-to-analog converter.

[0021] Application of a brain-inspired circuit with context-dependent memory and state-dependent memory, application in industrial robot damage detection. This application detects the state changes of the robot in the working environment and uses the context-dependent memory and state-dependent memory models to store and extract damage information.

[0022] The present invention provides a brain-inspired circuit with context-dependent memory and state-dependent memory and its application. It has the following beneficial effects:

[0023] The present invention simulates the processes of information encoding, storage, and retrieval in the form of a brain-like circuit, and combines a memristor matrix and a neuromorphic computing unit to achieve dynamic storage and extraction of information based on context-dependent memory and state-dependent memory. Compared with traditional damage detection techniques, the adaptive damage memory and extraction mechanism of the present invention can effectively reduce false alarms and missed alarms, improve the accuracy of industrial robot damage assessment, and can meet the requirements of multi-variable damage monitoring in complex production environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a circuit structure diagram designed based on the three-level processing model of memory information in the embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of the information memory and retrieval circuit in the embodiment of the present invention;

[0026] Figure 3 It is the operation result of the information memory, retrieval, and reset processes in the embodiment of the present invention;

[0027] Figure 4 It is the operation result of context-dependent memory promoting information retrieval in the embodiment of the present invention;

[0028] Figure 5This is the circuit structure diagram for detecting the damage degree of an industrial robot in the embodiments of the present invention; Specific embodiments

[0029] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] The embodiments of the present invention provide a brain-inspired circuit with context-dependent memory and state-dependent memory, including:

[0031] A sensory memory module for receiving information stimuli and encoding them to obtain information encoding information;

[0032] A short-term memory module for receiving the information encoding information output by the sensory memory module and deciding whether to repeat and store the information encoding information in long-term memory;

[0033] A state module for receiving psychological state information and encoding it to obtain state encoding information;

[0034] A context module for receiving context information and encoding it to obtain context encoding information;

[0035] A long-term memory module for receiving the information encoding information repeated by the short-term memory module and storing the state encoding information of the state module and the context encoding information of the context module in long-term memory;

[0036] An information retrieval module for receiving an information retrieval signal and extracting corresponding stored information from the long-term memory module according to the clue signals provided by the context and / or state, and transmitting the stored information to the short-term memory module.

[0037] Specifically, the brain-inspired circuit works through the cooperation of multiple modules to simulate the context-dependent memory and state-dependent memory mechanisms of the biological brain, and realizes efficient and low-power information storage and extraction through a memristor circuit.

[0038] The sensory memory module is used to receive external information stimuli and encode and process them. It includes at least one analog sensor that converts external environmental signals into electrical signals, which are then converted into digital signals by an analog-to-digital converter (ADC) and input into the short-term memory module in the form of information-encoded information. The short-term memory module is mainly composed of a CMOS logic circuit and a programmable memristor matrix. It receives the output of the sensory memory module and adjusts the resistance value of the memristors according to the frequency and intensity of the input signals, thereby realizing the storage of information-encoded information. When the intensity or repetition frequency of the input signal exceeds the set threshold, this module transmits the information-encoded information to the long-term memory module; otherwise, the information-encoded information will decay and be forgotten over time within the short-term memory module.

[0039] The long-term memory module consists of an information long-term memory sub-module, a context long-term memory sub-module, and a state long-term memory sub-module. Among them, the information long-term memory sub-module uses a memristor array to store the main information data; the context long-term memory sub-module collects the external environmental parameters at the time of information storage through multiple environmental sensors (such as temperature and humidity sensors, light sensors, etc.) and performs feature encoding and stores them in the memristor matrix for subsequent information retrieval; the state long-term memory sub-module is used to store the mental state information at the time of information storage. This module uses a neuromorphic circuit to simulate the state-dependent characteristics of biological neural networks, so that corresponding memory storage is more easily activated under specific mental states.

[0040] The information retrieval module performs matching operations on the stored content through a neuromorphic computing unit and a variable-weight memristor matrix. This module receives external information retrieval signals and combines the cue signals (VCue) provided by the context module and the state module to determine the retrieval priority of the target information. When the context matching signal provided by the context module is relatively high, or the state matching signal provided by the state module is relatively high, the information retrieval module will strengthen the access to the information-encoded information, context-encoded information, and state-encoded information in the long-term memory module and transfer them into the short-term memory module for information acquisition cycling.

[0041] The state module consists of a synapse-like plasticity circuit that is used to calculate the matching degree between the current mental state and the stored state and output a state-encoded information cue signal; this module analyzes the electrical signals output by the neuromorphic circuit and combines the state-encoded information stored in the memristors to determine whether the current mental state matches the state at the time of storage and sends a weighted signal to the information retrieval module to enhance the extraction probability of relevant information.

[0042] The situation module consists of an environmental information processing circuit and a memristor array. This module receives current environmental perception data, performs matching calculations with the stored situation coding information, calculates the matching degree using a memristor matrix based on a fuzzy matching algorithm, and generates a situation coding information clue signal, thereby enhancing the retrieval effect of relevant memories under similar environmental conditions.

[0043] Based on the above technical solution, it will be applied to the implementation of industrial robot damage detection below, and a circuit structure will be designed based on the three-level processing model of memory information:

[0044] First, the information processing of the brain is dynamic and complex, usually involving multiple brain regions. The hypothalamus receives external environmental stimuli. The amygdala is an important area for generating positive and negative emotions based on external environmental stimuli. The hippocampus retains and processes various memories and plays an important role in learning and memory. The three-level processing model of memory information describes the information processing process. When external information enters the memory system, it first enters sensory memory, and only those sensory information that attracts the individual's attention will enter short-term memory. The information coding information stored in short-term memory is rehearsed and stored in long-term memory. The information coding information stored in long-term memory is approximately invariant or permanent and will be retrieved into short-term memory when needed. In addition, when an organism memorizes information, situation coding information and mental state coding information are transferred to long-term memory for storage and retrieval. Inspired by the above mechanism, a circuit structure diagram based on the three-level processing model of memory information and with situation-dependent memory and state-dependent memory is designed, as Figure 1 shown. The environmental module performs the function of the thalamus, and the state module performs the function of the amygdala. The sensory memory module, short-term memory module, long-term memory module, and information retrieval module can realize the three-level processing of information, thereby performing the function of the hippocampus. Among them, the long-term memory module is divided into three types: the long-term memory module of information, the long-term memory module of the environment, and the long-term memory module of the state.

[0045] As Figure 2 This is a schematic diagram of the circuit for memorizing and retrieving information in two situations and two states in the embodiment of the present invention. The entire circuit is divided into a sensory memory module, a short-term memory module, a long-term memory module, an information retrieval module, a state module, and an environmental module. This circuit realizes the sensory memory, short-term memory, and long-term memory of information and considers various factors affecting information extraction. The detailed design of each module is as follows.

[0046] Sensory memory module: The sensory memory module is responsible for stimulating the information coding information and can automatically ignore weak information stimuli. When there is no information stimulus, the sensory memory module is in an inhibitory state, and V Sens has no output. Otherwise, the capacitor C1 is charged through the resistor R1. When the voltage V C1When it exceeds the threshold voltage of comparator COMP1, which is 1.4V, V Sens is at a high level. If the information stimulus is removed, P1 conducts, and the capacitor C1 discharges through resistors R1 and R2. For low-intensity pulse signals, this module can automatically ignore them.

[0047] Short-term memory module: Attention is the gateway to memory. Only when we notice certain information will this information enter short-term memory. Inspired by the above physiological mechanism, a short-term memory module is designed. Among them, the output signal V Sens of the sensory memory module serves as the input signal of the short-term memory module. V R represents the signal indicating successful information encoding and extraction. The attention signal V Att plays a crucial role in regulating the learning and forgetting rates of short-term memory. Assume V Att ∈[0V, 2V]. A higher V Att represents a stronger attention intensity. The basic functions of the short-term memory module are explained as follows:

[0048] Due to the existence of COMP2, if V Att is less than 0.5V, the switch S2 is opened, and V S2 is at a low level of 0V. Otherwise, the switch S2 is closed, and when V Sens or V R is activated, V S2 is the learning voltage V MR . It mimics the phenomenon that organisms have the ability of selection and self-determination and has the ability to distinguish between key and non-essential information. V Att regulates the learning rate and forgetting rate of organism information memory through the voltage summation function of OP1 and OP2. The voltage V OP2 is expressed as:

[0049] V OP2 =(0.1×V Att +V S2 +V MR );

[0050] When V S2 is the learning voltage V MR , the voltage V OP2 causes the resistance value of the memristor M S to decrease. If V MS is greater than 0.43V, COMP3 outputs a high level. V Sens and V R generate V Re of -0.7V and V Re of -0.6V respectively during the period when COMP3 outputs a high level. Among them, V Re is the rehearsal signal for conversion to long-term memory.

[0051] Long - term memory module: The long - term memory module is responsible for retaining long - term memories of information, situations, and states. Therefore, the long - term memory module is divided into three types. Among them, the circuit structure of the long - term memory module for situations is the same as that of the long - term memory module for states, and the circuit structure of the long - term memory module for information is similar to that of the long - term memory module for states. Taking the long - term memory module 2 for states as an example, its circuit structure is analyzed. V Re is controlled by signal V Ctrl . V Read is the information reading signal, and this voltage is less than the threshold of the memristor M L , and it will not cause a change in its resistance value. The cue signal V Cue3 is the output signal of this module. When encoding the mental state information into long - term memory, V Ctel3 makes N7 conduct, and V Re is input into the adder, and the resistance value of the memristor MST2 decreases. When extracting information, the long - term memory module outputs the cue signal V Cue .

[0052] Meanwhile, this module adds a reset function, which brings flexibility and configurability to information storage. When V Reset is at a high level, a voltage of - 5.2V acts on the memristor M L . The memristor M L is forced to return to the high - resistance state to receive new information.

[0053] Information retrieval module: The information retrieval module determines whether information can be retrieved. V Ext is the information retrieval signal. V Re is the recall signal, representing the signal of successful information retrieval. The cue signal V Cue1 comes from the long - term memory module for information, the cue signals V Cue2 and V Cue3 come from the long - term memory modules for different states, and the cue signals V Cue4 ~V CueN come from the long - term memory modules for different situations. These cue signals, as the input signals of the information retrieval module, all act on the inverting input terminal of OP4. V OP4 It is expressed by the formula as:

[0054]

[0055] where j = 2, 3; i = 4, 5, 6…. If the mood state during information retrieval is the same as the mood state during memory information, one of the cue signals in 4。 acts on OP If the state encoding information during information retrieval is the same as or similar to the situation encoding information during memory information,A certain clue signal in it acts on OP4. Recall signal V R serves as a signal indicating successful information retrieval, and when V C2 exceeds 0.3V. When V OP4 exceeds 1.5V, switch S3 closes, and V OP4 is applied to the integrated RC circuit. The higher the voltage of V OP4 , the faster the capacitor C2 charges, which means the voltage V C2 rises faster. The output voltage V Sens of the sensory memory module can also cause a voltage of 1V to be applied to the RC integrated circuit. The faster the voltage V C2 reaches 0.3V, the earlier the signal V R will be generated. Note that when the signal V R is generated, a signal V of -0.6V Re will further cause V C2 to rise rapidly. Compared with after the signal V Sens disappears, the voltage V C2 exceeds 0.3V for a longer time after V R disappears. This conforms to the actual situation because the learning effect obtained through recall is better than that obtained through memory.

[0056] State module: The state module can generate two types of state-encoded information, happiness and sadness, according to different types of environmental stimuli. Taking the generation and disappearance of the happy state as an example, the state module is introduced. V CTX represents the situation voltage. When the situation stimulus exceeds 6V, V P is 1.4V, which means the individual is exposed to a positive environment. A voltage of 1.4V is applied to the memristor M P , and its memristance gradually decreases. When V MP exceeds 0.43V, COMP5 outputs a high level, indicating the generation of the happy state. If the situation stimulus exceeding 6V is removed at this time, COMP5 changes from a high level to a low level, and V P becomes the low level of -1.4V. -1.4V is applied to the memristor M P , and its memristance gradually increases. When the voltage V MP is less than -0.4V, COMP4 will output a high level. When the output signal V Ctrl of the OR gate OR3 is at a high level, it indicates the existence of the happy state-encoded information. The sad state-encoded information is similar to the happy state-encoded information. Only when the situation voltage is less than 4V, it may cause the generation of the sad state-encoded information.

[0057] Situation Module: The situation module can achieve the memory of the situation where information is memorized and retrieved. Taking situation module 1 as an example, its circuit structure is introduced. The voltage follower OP8 has a large input resistance, which can not only follow the change of voltage V C3 but also prevent the loss of the charge stored in C3. Therefore, the initial voltage of V OP8 is 0V. When the signal V Con is input for the first time, A ND5 outputs a high level, and the switch S5 closes, which makes V CTX charge the capacitor C3 through D2 and R2. Due to the small parameters of the resistor R2 and the capacitor C3, V C3 rises to V CTX in a very short time. When the voltage V OP8 exceeds 1.5V, -2.2V is applied to the memristor M A , and its memristance decreases. V A changes from a positive voltage to a negative voltage. When V A is less than -0.3V, COMP8 outputs a high level, and the output of NOT1 becomes a low level. Therefore, the switch S5 disconnects, and the circuit forms a self-lock, preventing the situation voltage from entering this module. If the reset signal V Reset4 is input, C3 is immediately discharged, and the output of NOT1 becomes a high level, which allows this module to store a new situation voltage again. In the situation voltage comparator, the addition and subtraction operation circuit consists of an amplifier and resistors. 0.5V is applied to the inverting input terminal of the amplifier to compensate for the voltage drop of the diode D2. For the addition and subtraction operation circuit with three input signals, by using the superposition principle to select appropriate resistance values to achieve the following operation relationship:

[0058] V O = 2×(V CTX - V IN - 0.5V);

[0059] The output voltage of the amplifier (V O ) is twice the input voltage difference. When V0 is -0.7V to 0.7V, the situation voltage stored in this module differs from the situation voltage input again by -0.35V to 0.35V. At this time, V OUT is at a high level, indicating that the stored situation is the same as or similar to the situation input again. When V Con is input again, if the situation stored in situation module 1 is not similar to the situation input again, then V STO1 is at a high level. V STO1 allows situation module 2 to store the re-input situation. Similarly, when V STONWhen it remains high for a period of time, the situation module N+1 stores new situations. At the same time, the situation module is scalable. If this module and the long-term memory module of the situation are regarded as a whole and matched into the circuit, multi-situation memory can be achieved.

[0060] Figure 3 It is the operation result of the information memory, retrieval and reset processes. The results are divided into three stages, and each stage is not affected by context-dependent memory and state-dependent memory. For stages 1, 2, and 3, the time intervals between information memory and information extraction are 1.74 s, 1.16 s, and 0.76 s respectively. And the operation results are carried out under the condition of V Att = 1.2 V.

[0061] Stage 1: During 0 - 0.26 s, the input information encodes the information stimulus V S , and the short-term memory is enhanced. When the memristance value of M S drops below 6.5 kΩ, a signal V Re of -0.7 V is generated and applied to the long-term memory module of the information encoding information. The memristance value of M L drops, but its resistance value exceeds 8 kΩ at 0.26 s. This causes V Ext not to reach 0.3 V when the retrieval signal V C2 is input. Therefore, during 2 - 2.86 s, the recall signal V R is not generated and the retrieval of information cannot be completed. During 4 - 4.26 s and 8 - 8.26 s, the memristance of M L drops to 7.76 kΩ and 7.40 kΩ respectively. During 6 - 6.86 s and 10 - 10.86 s, V Ext is input in sequence, and the retrieval processes take 0.307 s and 0.254 s respectively, which shows that the more deeply the information encoding information is memorized, the shorter the time taken for the retrieval process. During the retrieval, the -0.6 V V R generated by the recall signal can consolidate the memory of the information. During 12.6 - 13.46 s, V Re is input again. Note that this retrieval is different from the previous one, and its time interval from the previous information retrieval is 1.74 s. The V Ext generated during the previous information retrieval causes V Re to rise rapidly, which results in a longer time for the voltage V C2 to drop to 0.3 V after V Ext disappears. Therefore, during 12.6 - 13.46 s, the time taken for the retrieval is affected by V C2 and M C2 and M LThe influence on the memristance values of both is such that the retrieval time is shortened, and the time taken for retrieval is only 0.128 s. This is consistent with the fact that the memory effect is enhanced through information retrieval.

[0062] Phase 2: During 15.5 - 15.7 s, the input signal V Reset1 , the resistance value of the memristor M L returns to the high resistance state to receive information again. From 15.8 - 16.84 s, the input information stimulates VS, and the memristance value of M L drops to 7.39 kΩ. During 18 - 18.86 s, with the input of V Ext , the retrieval process takes 0.248 s.

[0063] Phase 3: Phase 3 is similar to Phase 2, and the retrieval process in Phase 3 takes 0.191 s. By comparing the retrieval times of Phase 2 and Phase 3, it shows that after information is memorized, the earlier it is retrieved, the shorter the time taken for information retrieval.

[0064] Figure 4 This is the operating result of context - dependent memory promoting information encoding and information retrieval. Among them, Phase 1 is similar to Phase 2, except that Phase 1 has a context - dependent effect. Process 3 is the process of resetting the context of memory and then being able to memorize a new context again. It should be noted that Figure 4 In each stage, the time interval between information memorization and information extraction, as well as the attention, remain unchanged.

[0065] Phase 1: Process 1: During 0 - 0.44 s, the input signal VS deepens short - term memory. When MS drops below 6.5 kΩ, a consolidation signal VCon is generated. This signal causes the voltage VC2 to rise rapidly. It should be noted that the voltages reached by VC3 and VC4 differ from the actual context voltage by 0.5 V. The difference in voltage is compensated in the context memory module. During the information memorization stage, the resistance values of the memristors ML and MCTX1 drop, where ML drops to 7.73 kΩ. During 2.5 - 3.06 s, the context voltage is 4.7 V. This voltage differs from the context voltage (4.5 V) during the information memorization stage by 0.2 V and is considered a similar context by the circuit. Therefore, during the information extraction stage, the signal VOut4 remains high - level all the time, which enables the cue signal VCue4 to have a voltage that promotes information extraction. The retrieval process takes 0.253 s. During the information extraction stage, the generated signal VRe can consolidate the memory of the context and information again. During 3.5 - 3.7 s, the information is reset.

[0066] Stage 2: During 5 - 5.44 s, for the input signal VS, the resistance of the memristor ML still drops to 7.73 KΩ. Since the context voltage at this time (4.6 V) is similar to that in Stage 1, when memorizing information, the resistance of the memristor MCTX1 continues to decrease. During 7.5 - 8.06 s, the context voltage is 5.8 V. This voltage differs from the context voltage (4.6 V) during information memorization by 1.2 V and is considered a different context by the circuit. Therefore, at the initial stage of the information extraction phase, VOut3 is at a low level and the cue signal VCue4 does not have a voltage that promotes extraction. The retrieval time of Process 2 is extended to 0.307 s compared to Process 1. After successful retrieval, the generated signal VRe can cause the voltage VC3 to rise to the context voltage at that time, and the resistance of the memristor MCTX2 will decrease. This represents the formation of a memory of the new environment.

[0067] Stage 3: This stage demonstrates the flexibility and configurability of the context module. During 8.2 - 8.4 s, the input signal resets VReset4, the memristor MCTX1 returns to the high - resistance state, and the voltage VC3 drops to 0 V. With the context voltage at 3 V, the input stimulus signal VS is applied. VC2 rises to 2.5 V, and the resistances of the memristors ML and MCTX1 decrease, indicating the formation of a memory of the new environment (3 V). In addition, after applying the input reset signal VReset5, the signal Vs also causes the resistances of the memristors ML and MCTX2 to decrease. This shows that the circuit can simultaneously memorize two different contexts. By integrating the context module and the long - term memory module of the context as a whole into the circuit, multi - context memory can be achieved.

[0068] Figure 5 The circuit structure diagram for detecting the damage degree of an industrial robot. Industrial robots are widely used in the field of automobile manufacturing. By replacing the end - effector, industrial robots can operate in working environments such as welding, spraying, assembly, and handling. However, when industrial robots operate in different working environments, they are affected by various environmental factors. These factors may affect the performance, accuracy, lifespan, and even safety of the robot. For example, high temperature may cause the robot's motor to overheat, lubricating oil to fail, and the performance of electronic components to decline. Low temperature may cause mechanical components to become brittle, lubricants to solidify, and battery performance to decline. High humidity may cause electrical components to short - circuit, metal parts to rust, and sensors to fail. Dryness may generate static electricity, causing wear of precision parts. Vibration may cause the robot's positioning accuracy to decline, screws to loosen, and components to wear.

[0069] The present invention extends the proposed brain - like circuit and designs a circuit for detecting the damage degree of industrial robots in the field of automobile manufacturing, such as Figure 5As shown. The sensor detects the working environment of the industrial robot. If there are environmental factors that damage the industrial robot, the status module will generate two signals. These two signals are converted into damage voltage V in the receptor. Dmge The sensory memory module can receive V Dmge and resist signal interference. The sensory memory module is modulated by the signal V Att and outputs voltages V Con and V Re . The higher the voltage of V Att , the weaker the ability of the industrial robot to resist abnormal environments. The situation module 1 receives V Con and can store the working environment information when the industrial robot is damaged for the first time. The situation module 2 receives the voltage V STO1 and can store the working environment information when the industrial robot is damaged for the second time. By inputting the retrieval signal V Ext , the long-term memory module can visually display the specific situation of the industrial robot damage. According to the retrieval result, appropriate measures are taken, which can reduce the subsequent damage to the robot to a certain extent. By inputting the reset signal, the information stored in the circuit is initialized to receive the information about the industrial robot damage again. Figure 5 The circuit in has scalability. If the situation module and the long-term memory module of the situation are regarded as a whole and matched into the circuit, it can remember the damage suffered by the industrial robot in various working environments. It should be noted that the information obtained by the circuit is stored in the memristor. This means that even if an accidental power outage occurs during operation, the data will not be affected.

[0070] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A brain-inspired circuit with context-dependent memory and state-dependent memory, characterized in that, Including: A sensory memory module, configured to receive information stimuli and encode them to obtain information encoding information; A short-term memory module, configured to receive the information encoding information output by the sensory memory module and determine whether to repeat and store the information encoding information in long-term memory; A state module, configured to receive psychological state information and encode it to obtain state encoding information; A context module, configured to receive context information and encode it to obtain context encoding information; A long-term memory module, configured to receive the information encoding information repeated by the short-term memory module, and perform long-term memory storage on the state encoding information of the state module and the context encoding information of the context module; An information retrieval module, configured to receive an information retrieval signal and extract corresponding stored information from the long-term memory module according to the clue signals provided by the context and / or state, and transmit the stored information to the short-term memory module.

2. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, characterized in that, The output of the short-term memory module is regulated by an attention signal, and this attention signal controls the information learning rate and forgetting rate.

3. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, wherein The long-term memory module includes an information long-term memory module, a context long-term memory module, and a state long-term memory module. The information long-term memory module is used to store information encoding information, the context long-term memory module is used to store context encoding information, and the state long-term memory module is used to store state encoding information. And the stored contents of the information long-term memory module, the context long-term memory module, and the state long-term memory module cooperate with each other during information retrieval to improve the success rate of information extraction.

4. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, characterized in that The context module and the state module analyze the current environment or psychological state, match them with the stored context encoding information or state encoding information to output clue signals, which are used to accelerate information extraction and dynamically adjust the success rate of information retrieval according to the matching degree of the context or state.

5. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, characterized in that, The information retrieval module receives the information retrieval signal and the clue signal and determines whether to accelerate information recall. If the context or state matches, the clue signal is enhanced.

6. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, characterized in that, The context module analyzes the information in the current environment and matches it with the stored context encoding information, thereby providing a clue signal for information extraction.

7. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, characterized in that, The state module receives the signal of the current psychological state, compares it with the stored state encoding information, and provides a clue signal during information retrieval to accelerate the recall process.

8. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 6, characterized in that The information includes temperature, humidity, time, and location.

9. The brain-inspired circuit with context-dependent memory and state-dependent memory according to claim 1, characterized in that, The long-term memory module includes a calculation and feedback module, which is configured to calculate the matching degree between the current context encoding information and state encoding information and the stored data through a multiply-accumulate operation circuit, adjust the weight value of the memristor matrix according to the matching result, and provide a feedback signal to the information retrieval module through a digital-to-analog converter.

10. Application of a brain-inspired circuit with context-dependent memory and state-dependent memory. The brain-inspired circuit with context-dependent memory and state-dependent memory according to any one of claims 1-9, characterized in that, Application in industrial robot damage detection, which detects the state change of the robot in the working environment and uses the context-dependent memory and state-dependent memory models to store and extract damage information.

Images