Method and system for detecting faults in an electrolytic device having electrolytic cells

Abstract

The present invention provides a method and system for detecting malfunctions in an electrolytic device having an electrolytic cell. [Solution] A method, system, and assembly for detecting a fault in an electrolytic device having multiple electrolytic cells. The method includes: obtaining voltage measurements of electrolytic cells during operation of the electrolytic device; generating a composite cell voltage for the electrolytic cells using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters; comparing the voltage measurements with the composite cell voltage for the corresponding electrolytic cells to obtain a voltage difference; and detecting a fault in the electrolytic device when at least one of the voltage differences reaches a threshold.

Description

[Technical Field]

[0001] This disclosure relates in general to electrolytic devices, and more specifically to fault detection within electrolytic device cells. [Background technology]

[0002] Chloride-alkali electrolysis is a process that decomposes relatively low-value chemicals (e.g., NaCl, KCl) into higher-value chemicals (e.g., NaOH, Cl2, KOH) by applying a direct current. This reaction takes place in an electrochemical cell. In industrial settings, multiple cells are combined in series or parallel to achieve the reaction. This combination is called an electrolytic apparatus.

[0003] A chlorine-alkali electrochemical cell consists of an anode, cathode, and separator. Oxidation reactions occur at the anode, and reduction reactions occur at the cathode. In some cases, ion-exchange membranes can be used to separate the cathode and anode reactions. For a chlorine-alkali electrochemical cell, the primary electrolytic products are chlorine, hydrogen, and sodium hydroxide or potassium hydroxide, also known as "caustic." [Overview of the project] [Problems that the invention aims to solve]

[0004] There is a need for improved methods to detect faults within electrolytic cell cells. [Means for solving the problem]

[0005] In a broader embodiment, a method is provided for detecting a fault in an electrolytic apparatus having multiple electrolytic cells. The method includes: obtaining voltage measurements of the electrolytic cells during operation of the electrolytic apparatus; generating a synthetic cell voltage for the electrolytic cells using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters; comparing the synthetic cell voltage with the voltage measurement for a corresponding electrolytic cell among the electrolytic cells to obtain a voltage difference; and detecting a fault in the electrolytic apparatus when at least one of the voltage differences reaches a threshold.

[0006] In another broader embodiment, a system is provided for detecting faults in an electrolytic apparatus having multiple electrolytic cells. The system includes a processing unit and a non-temporary computer-readable medium storing program code thereon. The program code is executable by the processing unit to obtain voltage measurements of the electrolytic cells during the operation of the electrolytic apparatus, generate a composite cell voltage for the electrolytic cells using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters, compare the composite cell voltage with the voltage measurement for a corresponding electrolytic cell among the electrolytic cells to obtain a voltage difference, and detect a fault in the electrolytic apparatus if at least one of the voltage differences reaches a threshold.

[0007] In yet another broader embodiment, an assembly is provided comprising an electrolytic apparatus having a plurality of electrolytic cells and at least one computing device operationally coupled to the electrolytic apparatus. The at least one computing device includes at least one processing unit and a non-temporary computer-readable medium storing program instructions thereon. The program instructions are executable by at least one processing unit to obtain voltage measurements of the electrolytic cells during the operation of the electrolytic apparatus, generate a composite cell voltage for the electrolytic cells using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters, compare the composite cell voltage with the voltage measurement for a corresponding electrolytic cell among the electrolytic cells to obtain a voltage difference, and detect a fault in the electrolytic apparatus if at least one of the voltage differences reaches a threshold.

[0008] The features of the systems, devices, and methods described herein can be used in various combinations according to the embodiments described herein.

[0009] Please refer to the attached diagram below. [Brief explanation of the drawing]

[0010] [Figure 1] This is a schematic diagram of an exemplary electrolytic cell. [Figure 2] This is a block diagram of an exemplary assembly for detecting faults within an electrolytic device. [Figure 3] This is a block diagram of an exemplary embodiment of a fault detection system. [Figure 4] This is a flowchart illustrating an exemplary method for detecting a malfunction in an electrolytic device. [Figure 5] This is a block diagram of an exemplary computer device.

[0011] It should be noted that throughout the attached drawings, similar features are identified by the same reference number. [Modes for carrying out the invention]

[0012] Chloride-alkali production is an energy-consuming process, and therefore the efficiency of each electrochemical cell is a matter to consider during operation. Another issue to consider in the operation of chloride-alkali electrochemical cells is the prevention of hazards such as gas release, fire, and liquid leakage. This specification describes a method and system for the early detection of faults occurring in a chloride-alkali electrolytic cell operating in series with other cells in an electrolytic apparatus. The method involves acquiring a voltage and processing cell measurements that represent healthy behavior. A neural network model is constructed based on an encoder-decoder architecture, and the cell voltage is predicted using this neural network model. The predicted voltage is compared to the measured voltage for the purpose of detecting abnormal deviations while taking into account normal cell degradation.

[0013] Figure 1 is a schematic representation of an exemplary membrane cell 100 used in the chlor-alkali industry. It may consist of two compartments. In the case of salt electrolysis (NaCl), a saturated brine solution (NaCl) is packed into the anode compartment 102, while diluted caustic soda passes through the cathode compartment 103. In a chlor-alkali plant, chlorine (Cl2) 104 is produced at a coated (e.g., titanium (Ti)) anode 105. The combination of moving sodium ions 107 and hydroxide ions 106 traversing the selective membrane 108 produces caustic soda (NaOH) and hydrogen gas 109. The cathode 110 may be nickel with a catalytic coating to reduce overpotential due to hydrogen (H2) accumulation.

[0014] Voltage fluctuations within the membrane cell 100 are the result of physical changes within the cell components. The cell voltage drop is distributed among its components, namely the anode 105, cathode 110, membrane 108, and electrical connections. A cell voltage drop or rise can be considered as a precursor to two types of degradation: (1) normal degradation and end-of-cell life, and (2) abnormal / sudden degradation during its life. The cell may degrade abnormally because the membrane 108 malfunction allows the two compartments 104, 109 to not be properly separated, thus enabling unwanted chemical reactions. The cell may also degrade abnormally due to the electrodes 105, 110 losing their activation coatings. The cell may also degrade abnormally due to a combination of a malfunctioning membrane 108 and electrodes 105, 11, The root cause of component malfunction can be due to poor (external) operating conditions of the electrolyzer or the cell itself. These conditions include, but are not limited to, insufficient inlet flow rate, poor control of inlet contaminants, electrical hazards, poor temperature control, poor equipment assembly, etc.

[0015] A synthetic voltage is calculated for one or more of the cells 100 within a chlor-alkali electrolyzer to detect early abnormal degradation of the cell. This synthetic voltage is a calculation of a mathematical function composed of many estimated parameters. Some of these parameters are the same for all cells performing electrolysis under the same operating conditions and are operation-specific. Some parameters are cell-specific parameters. The calculation may take into account the normal degradation of the cell 100 that changes over time. An exemplary mathematical function is described below: <'

Equation

[0017] During operation, the individual cell voltages of the electrolyzer can be monitored to detect abnormal variations that could lead to significant safety issues. However, if only abnormal degradation is considered, a fault is detected when the cell has reached an irreparable point and needs to be replaced. To avoid this problem, normal cell degradation is also taken into account.

[0018] The voltage drop of cell 100 follows Ohm's law, and thus the relationship between the supply current and the cell voltage is directly proportional. Using the theoretical voltage estimated based on this proportionality, a healthy cell can be characterized. However, characterizing a cell using only current-voltage proportionality does not take into account the root and complex effects occurring within the anode and cathode compartments 104, 109. Therefore, multiple other parameters such as cell specificities due to age, technology, and location are also considered.

[0019] For the same operating conditions and normal degradation levels, the voltage of a given cell may differ from the voltages of other cells within the same electrolyzer. This difference can be induced by disparities in manufacturing, installation, and other factors that are difficult to identify. Therefore, cell-specific parameters such as age, technology, and location can be used.

[0020] The voltage drift of a cell can occur slowly over time due to cell degradation. Additionally, there may be a delay between a change in the operating conditions of the electrolyzer and the cell's response. To address this delay, the operating conditions at the preceding time step can be considered by a prediction model.

[0021] For the purpose of predicting abnormal degradation early, the predicted voltage is independent of the measured voltage. Therefore, the measured voltage is not used as an input to the prediction model.

[0022] In some embodiments, a neural network modeling architecture is used to calculate the combined voltage for the cells of an electrolytic device. This model is deployed in a production environment where the operating conditions are not predefined and each cell has a different typical degradation level. The neural network model may be based on an encoder-decoder architecture, where the decoder is replaced by predictors, which are subnetworks that predict the voltage of the cells. A neural encoder is a type of neural architecture that aims to take an input vector and reduce its dimensionality to a desired dimension. It can be paired with a decoder. The decoder receives the output of the encoder and transforms it to minimize an objective function.

[0023] Neural encoders can be used to discover features that represent cell specificities and, consequently, degradation during an operating cycle. Predictors can take into account the time delay of the measured values. While predictors do not use the measured voltage as input, they can still predict different voltages for each cell, even when using the same operating conditions as input. The predictor achieves this by taking the encoder output as a unique input for each cell. Therefore, the voltage prediction is not biased by the measured voltage of the cell.

[0024] Referring to Figure 2, an exemplary embodiment of assembly 200 including an electrolytic device 214 and a fault detection system 218 is shown. The fault detection system 218 comprises at least one computing device operationally coupled to the electrolytic device 214. In the exemplary embodiment, the fault detection system 218 includes a cell data acquisition device 252, a composite cell voltage generation device 272, and a fault detection device 262.

[0025] The cell data acquisition device 252 is configured to obtain voltage measurements of multiple electrolytic cells during the operation of the electrolytic device 214. Voltage measurements can be obtained in real time or pseudo-real time. To obtain voltage measurements, a single unit or multiple units can be connected to various cells. In some embodiments, voltage measurements are obtained for each cell of the electrolytic device 214.

[0026] The synthetic cell voltage generation device 272 is configured to generate a cell-specific synthetic cell voltage using a neural network architecture. The neural network takes into account normal cell degradation based on cell-specific parameters for the purpose of estimating the cell-specific synthetic cell voltage. In some embodiments, the neural network includes an encoder subnetwork and a predictor subnetwork. The encoder subnetwork may be configured to determine normal cell degradation based on cell-specific parameters. The predictor subnetwork may be configured to predict the synthetic cell voltage using the normal cell degradation output by the encoder subnetwork. In some embodiments, the predictor subnetwork is further configured to apply a time delay when predicting the synthetic cell voltage.

[0027] Various neural network architectures can be used to generate cell-specific synthesized cell voltages. For example, an encoder subnetwork may include a masking layer, a long short-term memory layer, and two dense layers, and output a two-dimensional vector representing cell-specific normal cell degradation. A predictor subnetwork may include two long short-term memory layers and two dense layers. Other embodiments are equally applicable depending on the practical implementation.

[0028] In some embodiments, the neural network is trained using historical data from the cells of the electrolytic device 214. In some embodiments, the neural network is trained using historical data from the cells of multiple electrolytic devices. Various techniques can be used to train the neural network to understand healthy cell behavior while taking into account normal cell degradation based on cell-specific parameters.

[0029] The fault detection device 262 compares the combined cell voltage from the combined cell voltage generation device 272 with the voltage measurement from the cell data acquisition device 252 for the corresponding electrolytic cell among the electrolytic cells in order to obtain a voltage difference. For example, a measured voltage and a combined voltage can be assigned to each electrolytic cell, and the difference between the measured voltage and the combined voltage corresponds to the voltage difference. In some embodiments, a voltage measurement is obtained for one subset of cells in the electrolytic apparatus 214. In some embodiments, voltage measurements are obtained alternately for multiple cell subsets. Other embodiments may be similar.

[0030] A fault is detected by the fault detection device 262 when one or more voltage differences reach a threshold. In some embodiments, fault detection includes issuing an alert signal, which may be an audible, visual, or other signal, to direct the attention of one or more operators to the fault. In some embodiments, fault detection includes issuing a shutdown signal to the electrolyzer 214 or to another device or system configured to shut down the electrolyzer 214. A combination of alert and shutdown signals may be used. For example, a first threshold may be associated with an alert signal, and a second threshold with a shutdown signal. If the voltage difference is within a first range, an alert signal is issued, and if the voltage difference is within a second range greater than the first range, a shutdown signal is issued. In another embodiment, the characteristics of the fault signal may be associated with the number of cells indicating a voltage difference that has reached a threshold. For example, if the voltage difference reaches a threshold for a first number of cells, an alert signal is issued, and if the voltage difference reaches a threshold for a second number of cells greater than the first number, a shutdown signal is issued. Other variations are similar.

[0031] Referring to Figure 3, a specific and non-limiting embodiment of the fault detection system 218 is shown. A data acquisition and transmission unit 202 measures the differential voltage of cathode-to-cathode or anode-to-anode cells in an electrolytic apparatus 214 with a given accuracy of + / - 1 millivolt or other precision level. The electrolytic apparatus 214 may be any available industrial chlorine-alkali electrolytic apparatus consisting of membrane cells connected in series. In some embodiments, the electrolytic apparatus 214 houses up to 160 cells, but the total number of cells may vary. Protected metal wires 213 connect the input terminal of the data acquisition and transmission unit 202 to the cathode or anode terminal in an adjacent cell. In some embodiments, each unit 202 may measure up to 32 power inputs, but the total number may vary. Unit 202 may house an analog-to-digital converter, a digital filter, a memory buffer and / or a microcontroller to perform acquisition and transmission routines. Unit 202 may be powered by a power supply 203 using protected metal wires 215.

[0032] Data originating from unit 202 can be transmitted to processing and communication unit 204. In addition to processing data transmission routines to the main computer server unit 209, unit 204 can execute and send emergency stop signals to the shutdown relay unit 206. Unit 204 may receive transformer-rectifier shunt current measurements from unit 205 using, for example, a 4-20mA converter terminal 217. Unit 204 may broadcast voltage and current data streams sampled at a given rate, e.g., one point per second, to unit 209. Ethernet communication unit 207 may broadcast process data streams originating from computing device 208 and not measured by units 202 and 204 to the computer server unit 209. Computing device 208 may be a third-party computer server, sometimes referred to as a distributed control system. Some exemplary process data include, but are not limited to, the catholite outlet temperature, outlet caustic concentration, and inlet and / or outlet pH. Units 202, 204, 209, 207, and 208 may be connected using optical fiber wire loops 216 or other connecting materials. In some embodiments, one or more of the connections are wireless. According to the illustrated embodiments, unit 209 is a main computer server that receives and processes voltage, current, and process data for a single cell, stores all data for one or more electrolyzers, performs a series of steps, and sends an electrolyzer shutdown command to unit 204 if present.

[0033] Unit 209 may be responsible for collecting and transforming historical data. A similar setup can be deployed in multiple electrolyzers 214 operating in multiple chlor-alkali plants. Unit 209 can store all voltage data, current measurements, and process data originating from each cell. Since the sampling rates of voltage measurements and process data may differ, Unit 209 can downsample voltage observations. In some embodiments, the collected process data includes outlet caustic concentration and electrolyzer outlet catholite temperature. Unit 109 can transform and sort the data into a tabular format where each row represents a different timestamp of the observation and each column represents a different measurement variable. Unit 109 can store all data in the database originating from several weeks of deployment of the fault detection system 218 from multiple electrolyzers.

[0034] Tabular data can be processed to select an operating cycle. An operating cycle is the time interval between consecutive starts and stops of an electrolytic device. Its length can range from a few hours to several weeks, depending on production constraints, changing demand, work shifts, or maintenance requirements. Each cycle can be divided into two phases of different lengths: a start phase and an operating phase. The start phase occurs when the current rises from zero to the permissible maximum value at which each cell reaches its maximum production condition. The rate at which the current rises may differ for each start due to changes in operating practices determined by the plant operator. The length of the start phase can vary, for example, from 20 minutes to 12 hours. The operating phase constitutes the remainder of the cycle. The current in this phase can vary, for example, between 50% and 100% of the full range.

[0035] In one specific and non-limiting example, the pseudocode for detecting possible cycles is as follows: For each observation n in the database (FOR): Diff n =Time n -Time n-1 Diffn If it is greater than 10 minutes (IF): Time n-1 and Time n are saved. Each possible cycle is (0, Time n-1 ), (Time n , Time n+1 ), (Time n+2 , Time n+3 ) ··· enclosed in parentheses of date pairs corresponding to.

[0036] In one specific and non - limiting example, the pseudo - code for checking cycles is as follows: For each possible cycle of length k n (FOR): If any observation result of the cycle has "current > 16kV" (IF): idx = the first observation result when "current > 16kV" If idx ≤ 12 hours (IF): startup n = cycle[0:idx] operation n = cycle[idx:k] If length(operation) ≥ length(startup) (IF): cycle n is a valid cycle.

[0037] Once the data is structured in cycles, a unity - based normalization scaling method can be applied to each of the data series. This linearly scales the data, and thus the values are within the range of [0, 1]. The scale helps to improve the training time for building a neural network model. The data series is scaled using the following equation: [Equation]

[0038] The minimum (min) and maximum (max) values ​​can be used to scale the measurement sequence.

[0039] In some embodiments, a neural network model is constructed and trained using collected and transformed historical data. The encoder subnetwork implies features that characterize the behavior of the cell voltage over a certain cycle during the startup phase. These features take into account both cell specificity and normal degradation. It is a self-supervising method as it does not require labeled degradation data. This step can be seen as performing dimensionality reduction. However, the encoded vector (dimensionality-reduced vector) is not the result of a statistical procedure, but rather an optimal representation that facilitates the training of the voltage predictor subnetwork. The length of each startup phase can vary and may be limited to a high value, such as up to 12 hours. For example, if each time step represents 1 minute, the vector given to this subnetwork has a length of 720 time steps. Three input features can be used: electrolytic cassolite outlet temperature, factory caustic concentration, and cell current and voltage. The input features can be used to normalize the voltage to the specific operating conditions of each startup. In some embodiments, the shape of the input vector is [720 time steps, 4 features]. A masking layer forces subsequent layers to ignore a time step if all features of that time step are equal to a masked value, where this masked value may be set to "-1" to filter out time steps that are added earlier during padding. To account for the transient nature of the sequence, the next layer may be a long short-term memory (LSTM). Then, two fully connected layers can be chained together to more smoothly transition to the final two-position encoded result. The output may be a vector of coordinates [X, Y] for each cell and the starting point, with the shape (1 time step, 2 features). Furthermore, these coordinates can be represented in a graph to provide insight into the decision process performed by the network.

[0040] A predictor subnetwork may be responsible for predicting the synthetic cell voltage. In some embodiments, the two inputs used for prediction are a window of time steps from the operating phase and an encoded representation of the cell's startup phase. According to this embodiment, a window of four observations is sufficient to represent the dynamics of the chemical phenomena behind the cell's response. The encoded cell startup can be iterated four times and coupled with a window of operating features. Two LSTM layers can be used to find the temporal correlation between the observations in each window. Two fully connected layers may follow to output the predicted voltage. The output layer may have a sigmoid activation function, as the output voltage may be pre-scaled to the range [0,1]. In this step, the entire model is trained by minimizing the loss between the predicted and measured voltages by this subnetwork. The Adam optimizer and backpropagation algorithm can be used. For this subnetwork to achieve good accuracy in voltage prediction, the encoder must learn a faithful representation of the cell's characterization.

[0041] The following pseudocode represents a specific and non-restrictive example of a training loop with one observation per forward-reverse pass: For each cycle: Define cell_startup and cell_operation. Separate the cell_operation within the window. For each window: operating_conditions=window(voltage excluded) cell_voltage = final voltage observation result within the window #Forward path encoded_cell_startup=encoder(cell_startup) predicted_voltage=predictor(operating_conditions,encoded_cell_startup) loss=mean_squared_error(predicted_voltage,cell_voltage) #backpropagation update_network_weights(loss)

[0042] To improve the computational efficiency of this training loop, mini-batch training can be used to parallelize the computation. In this training mode, many observations can be grouped into batches for parallel processing. According to this embodiment, one or more graphics processing units (GPUs) are used within unit 209 to perform this operation. Furthermore, three operations—padding, shuffling, and window striding—can be used to efficiently converge the network. Not all cycle starts have the same duration. However, all patches fed to the GPU must have the same number of time steps. This problem is addressed by padding the sequence, i.e., by adding a value of "-1" at the end of the corresponding start for each observation. In this way, all starts have the same duration of 720 minutes, which is equal to the maximum duration of a single start. This padding value is later ignored by the masking layer of the encoder subnetwork and therefore does not affect the result. To reduce the time required for the network to converge to the optimal solution, a shuffling procedure can be used in which each batch has observations from different cells, cycles, and electrolyzers. The task of the encoder, which implies the features of the cells being started, is even more complex than the task of the predictor. Therefore, it is more efficient to train the network with fewer observations and more different starting sequences per cycle. To address this problem, the stride of the window function can be increased. The stride is a number that defines how much of the sequence window to ignore between two consecutive training observations. At the end of the training loop, the constructed neural network predictors can be stored in a file, for example, in unit 209.

[0043] In some embodiments, the fault detection system 218 is deployed within one or more electrolytic devices in the factory. Voltages from multiple cells are measured, and process data may be acquired from third-party devices. Alignment and cycle detection routines may be executed in real time or pseudo-real time within unit 209. A pre-built neural network voltage predictor is used to determine the cell voltages.

number

number

[0044] Figure 4 is a flowchart of an exemplary method 400 for fault detection. In some embodiments, method 400 is performed by a fault detection system 218 of assembly 200 in Figure 2. In some embodiments, method 400 is performed by multiple computing devices.

[0045] In step 402, voltage measurements of the electrolytic cells are obtained. In step 404, a composite voltage for the electrolytic cells is generated, as described in more detail above. Steps 402 and 404 can be performed simultaneously. In some embodiments, step 404 can be performed before step 402. In step 404, the normal degradation of the electrolytic cells is taken into account by a neural network trained to learn the behavior of healthy electrolytic cells. The composite cell voltage is predicted using cell-specific parameters, and therefore the composite cell voltages of two or more cells in the same electrolytic device may be different.

[0046] In step 406, the voltage measurement obtained in step 402 is compared with the composite cell voltage generated in step 404 for the corresponding cell. In other words, the composite cell voltage of a given cell is compared with the measured cell voltage of the same cell. The comparison is performed cell by cell for any number of cells stored in the electrolysis. In some embodiments, the comparison is performed for all cells in the electrolysis, either simultaneously or sequentially, in any random or predetermined order. If one or more voltage differences obtained when comparing the composite voltage and the measured voltage for a given cell reach a threshold, a fault is detected in step 408.

[0047] Method 400 can be repeated any number of times at random or predetermined intervals. In some embodiments, Method 400 is performed continuously until at least one failure is detected. Other embodiments are similarly applicable.

[0048] Figure 5 shows an exemplary embodiment of a computing device 500 for implementing the method 400 for detecting faults in an electrolytic device as described above. In some embodiments, the fault detection system 218 is implemented using one or more computing devices 500. The computing device 500 includes a processing unit 502 and a memory 504 that stores computer-executable instructions 506 internally. The processing unit 502 may include any suitable device configured to perform a series of steps so that when the instruction 506 is executed by the computing device 500 or other programmable device, it can perform the function / action / step defined in the method 400 described herein. The processing unit 502 may include, for example, any type of general-purpose microprocessor or microcontroller, a digital signal processing (DSP) processor, a CPU, an integrated circuit, a field-programmable gate array (FPGA), a reconfigurable processor, other suitably programmed or programmable logic circuits, or any combination thereof.

[0049] Memory 504 may include any suitable known or other machine-readable storage medium. Memory 504 may include non-temporary computer-readable storage media, such as, without limitation, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Memory 504 may include any type of computer memory located either inside or outside the device, such as random access memory (RAM), read-only memory (ROM), electro-optical memory, magneto-optical memory, erasable programmable read-only memory (EPROM), and electrically erasable read-only memory (EEPROM), ferroelectric RAM (FRAM), and any suitable combination thereof. Memory 504 may include any storage means (e.g., devices) suitable for storing machine-readable instructions 506 executable by the processing unit 502 in a searchable format.

[0050] It should be noted that the technologies described herein can be implemented by computing device 500 in substantially real time.

[0051] The method and system for detecting faults in an electrolytic device as described herein may be implemented in a high-level procedural or object-oriented programming or scripting language or a combination thereof to communicate with or assist in the operation of a computer system, such as computing device 500. Alternatively, the method and system for detecting faults in an electrolytic device may be implemented in assembly language or machine language. The language may be a compiled language or an interpreted language. The program code for implementing the method and system for detecting faults in an electrolytic device may be stored on a storage medium or device, such as a ROM, magnetic disk, optical disk, flash drive or any other suitable storage medium or device. The program code may be readable by a general-purpose or dedicated programmable computer to configure and operate the computer when the storage medium or device is read by the computer to perform the procedures described herein. Embodiments of the method and system for detecting faults in an electrolytic device can similarly be considered to be implemented via a non-temporary computer-readable storage medium on which a computer program is stored. A computer program may include computer-readable instructions that cause a computer or, more specifically, a processing unit 502 of a computing device 500 to perform operations in a specific and predefined manner to perform the functions described herein.

[0052] Computer executable instructions can take many forms, including program modules executed by one or more computers or other devices. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. Typically, the functionality of program modules can be combined or distributed as desired in various embodiments.

[0053] The embodiments described herein provide non-limiting examples of possible implementations of the art. A person skilled in the art, having reviewed this disclosure, will recognize that modifications can be made to the embodiments described herein without departing from the scope of the art. For example, the software modules can be combined or separated in different ways for the purpose of performing the steps of Method 400, or the specific devices used to obtain various measurements from the electrolytic cell and / or electrolytic apparatus can vary. Furthermore, a person skilled in the art will likely be able to implement further modifications in consideration of this disclosure, and such modifications will likely fall within the scope of the art. This disclosure also discloses embodiments as illustrated below. [Embodiment 1] A method for detecting a malfunction in an electrolytic device having multiple electrolytic cells, During the operation of the electrolytic device, the voltage measurement value of the electrolytic cell is obtained; A neural network architecture is used to generate a composite cell voltage for the electrolytic cell, taking into account normal cell degradation based on cell-specific parameters; The voltage measurement is compared with the combined cell voltage of the corresponding electrolytic cell to obtain the voltage difference; A fault in the electrolytic device is detected when at least one of the aforementioned voltage differences reaches a threshold; A method that includes this. [Embodiment 2] The method according to Embodiment 1, wherein the neural network architecture includes an encoder subnetwork and a predictor subnetwork, the encoder subnetwork is configured to determine the normal cell degradation based on the cell-specific parameters, and the predictor subnetwork is configured to predict the composite cell voltage using the output of the normal cell degradation by the encoder subnetwork. [Embodiment 3] The method according to Embodiment 2, wherein the predictor subnetwork is further configured to apply a time delay when predicting the composite cell voltage. [Embodiment 4] The method according to Embodiment 2, wherein the encoder subnetwork includes a masking layer, a long short-term memory layer, and two fully connected layers, and outputs a two-dimensional vector representing the cell-specific normal cell degradation. [Embodiment 5] The method according to Embodiment 2, wherein the predictor subnetwork includes two long-short-term memory layers and two fully connected layers. [Embodiment 6] The method according to Embodiment 1, wherein detecting the fault in the electrolytic device includes outputting an alert signal. [Embodiment 7] The method according to Embodiment 1, wherein detecting the fault in the electrolytic device includes outputting a signal to shut down the electrolytic device. [Embodiment 8] The method according to Embodiment 1, wherein the neural network architecture is trained using historical data from multiple electrolytic devices. [Embodiment 9] The method according to Embodiment 1, wherein the electrolytic cell is a chlorine-alkali electrolytic cell. [Embodiment 10] A system for detecting faults in an electrolytic device having multiple electrolytic cells: Processing unit and; During the operation of the electrolytic device, the voltage measurement value of the electrolytic cell is obtained; A synthetic cell voltage for the electrolytic cell is generated using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters; The voltage measurement is compared with the combined cell voltage of the corresponding electrolytic cell to obtain the voltage difference; A fault in the electrolytic device is detected when at least one of the aforementioned voltage differences reaches a threshold; A non-temporary computer-readable medium storing program code executable by the processing unit; A system that includes this. [Embodiment 11] The system according to Embodiment 10, wherein the neural network architecture includes an encoder subnetwork and a predictor subnetwork, the encoder subnetwork is configured to determine the normal cell degradation based on the cell-specific parameters, and the predictor subnetwork is configured to predict the composite cell voltage using the normal cell degradation output by the encoder subnetwork. [Embodiment 12] The system according to embodiment 11, wherein the predictor subnetwork is further configured to apply a time delay when predicting the composite cell voltage. [Embodiment 13] The system according to Embodiment 11, wherein the encoder subnetwork includes a masking layer, a long short-term memory layer, and two fully connected layers, and outputs a two-dimensional vector representing the cell-specific normal cell degradation. [Embodiment 14] The system according to embodiment 11, wherein the predictor subnetwork includes two long-short-term memory layers and two fully connected layers. [Embodiment 15] The system according to embodiment 10, wherein detecting the fault in the electrolytic device includes outputting an alert signal. [Embodiment 16] The system according to Embodiment 10, wherein detecting the fault in the electrolytic device includes outputting a signal to shut down the electrolytic device. [Embodiment 17] The system according to embodiment 10, wherein the neural network architecture is trained using historical data from multiple electrolytic devices. [Embodiment 18] The system according to embodiment 10, wherein the electrolytic cell is a chlorine-alkali electrolytic cell. [Embodiment 19] Electrolytic apparatus having multiple electrolytic cells; and At least one computing device operatively coupled to the electrolytic device, An assembly that includes, The said at least one computing device comprises at least one processing unit and; During the operation of the electrolytic device, the voltage measurement value of the electrolytic cell is obtained; A synthetic cell voltage for the electrolytic cell is generated using a neural network architecture that takes into account normal cell degradation based on cell-specific parameters; The voltage measurement is compared with the combined cell voltage of the corresponding electrolytic cell to obtain the voltage difference; A fault in the electrolytic device is detected when at least one of the aforementioned voltage differences reaches a threshold; A non-temporary computer-readable medium storing program instructions thereon that can be executed by at least one processing unit, assembly. [Embodiment 20] The aforementioned at least one computing device: To obtain the voltage measurement value during the operation of the electrolytic device, a plurality of data acquisition and transmission units are coupled to the electrolytic cell; A processing and communication unit coupled to the aforementioned plurality of data acquisition and transmission units; A main computer server coupled to the aforementioned processing and communication unit; The assembly described in Embodiment 19, including the assembly described in Embodiment 19.

Claims

1. A method for detecting a malfunction in an electrolytic device having multiple electrolytic cells, During the operation of the electrolytic device, the voltage measurement value of the electrolytic cell is obtained; The method involves generating a composite cell voltage for an electrolytic cell using a neural network architecture trained with historical data from multiple electrolytic devices, wherein the historical data is composed of multiple operation cycles, each including a startup phase and an operation phase, and the neural network architecture is: An encoder subnetwork configured to determine cell degradation indicating the end of life for each electrolytic cell based on parameters representing the uniqueness of each electrolytic cell, which for each electrolytic cell includes at least one of the following: the number of years since the electrolytic cell, the technology of the electrolytic device, and the position of the electrolytic cell in the electrolytic device, and to output a two-dimensional vector representing the cell degradation at the start-up stage for each electrolytic cell, A predictor subnetwork configured to predict the composite cell voltage using a two-dimensional vector output by the encoder subnetwork, Including generating a composite cell voltage; The voltage measurement is compared with the combined cell voltage of the corresponding electrolytic cell among the plurality of electrolytic cells to obtain a voltage difference; A fault in the electrolytic device is detected when at least one of the aforementioned voltage differences reaches a threshold; including, method.

2. The method according to claim 1, wherein the encoder subnetwork includes a masking layer, a long short-term memory layer, and two fully connected layers.

3. The method according to claim 2, wherein the predictor subnetwork is further configured to apply a time delay when predicting the composite cell voltage.

4. The method according to claim 2, wherein the predictor subnetwork includes two long-short-term memory layers and two fully connected layers.

5. The method according to claim 1, wherein detecting the fault in the electrolytic device includes outputting an alert signal.

6. The method according to claim 1, wherein detecting the fault in the electrolytic device includes outputting a signal to shut down the electrolytic device.

7. The method according to claim 1, wherein the electrolytic cell is a chlorine-alkali electrolytic cell.

8. A system for detecting faults in an electrolytic device having multiple electrolytic cells, comprising: Processing unit and; During the operation of the electrolytic device, the voltage measurement value of the electrolytic cell is obtained; Using a neural network architecture trained with historical data from multiple electrolytic devices, a composite cell voltage is generated for the electrolytic cell; The voltage measurement is compared with the combined cell voltage for the corresponding electrolytic cell among the plurality of electrolytic cells to obtain the voltage difference; When at least one of the aforementioned voltage differences reaches a threshold, a fault in the electrolytic device is detected; A non-temporary computer-readable medium storing program code executable by the processing unit; Includes, The aforementioned historical data is composed of multiple operation cycles, each including a startup phase and an operation phase, and the neural network architecture is, An encoder subnetwork configured to determine cell degradation indicating the end of life for each electrolytic cell based on parameters representing the uniqueness of each electrolytic cell, which for each electrolytic cell includes at least one of the following: the number of years since the electrolytic cell, the technology of the electrolytic device, and the position of the electrolytic cell in the electrolytic device, and to output a two-dimensional vector representing the cell degradation at the start-up stage for each electrolytic cell, A predictor subnetwork configured to predict the composite cell voltage using a two-dimensional vector output by the encoder subnetwork, including, system.

9. The system according to claim 8, wherein the encoder subnetwork includes a masking layer, a long short-term memory layer, and two fully connected layers.

10. The system according to claim 9, wherein the predictor subnetwork is further configured to apply a time delay when predicting the composite cell voltage.

11. The system according to claim 9, wherein the predictor subnetwork includes two long-short-term memory layers and two fully connected layers.

12. The system according to claim 8, wherein detecting the fault in the electrolytic device includes outputting an alert signal.

13. The system according to claim 8, wherein detecting the fault in the electrolytic device includes outputting a signal to shut down the electrolytic device.

14. The system according to claim 8, wherein the electrolytic cell is a chlorine-alkali electrolytic cell.

15. Electrolytic apparatus having multiple electrolytic cells; and At least one computing device operatively coupled to the electrolytic device, An assembly that includes, The said at least one computing device comprises at least one processing unit; During the operation of the electrolytic device, the voltage measurement value of the electrolytic cell is obtained; Using a neural network architecture trained with historical data from multiple electrolytic devices, a composite cell voltage is generated for the electrolytic cell; The voltage measurement is compared with the combined cell voltage for the corresponding electrolytic cell among the plurality of electrolytic cells to obtain the voltage difference; When at least one of the aforementioned voltage differences reaches a threshold, a fault in the electrolytic device is detected; A non-temporary computer-readable medium storing program instructions that can be executed by at least one processing unit, Includes, The aforementioned historical data is composed of multiple operation cycles, each including a startup phase and an operation phase, and the neural network architecture is, An encoder subnetwork configured to determine cell degradation indicating the end of life for each electrolytic cell based on parameters representing the uniqueness of each electrolytic cell, which for each electrolytic cell includes at least one of the following: the number of years since the electrolytic cell, the technology of the electrolytic device, and the position of the electrolytic cell in the electrolytic device, and to output a two-dimensional vector representing the cell degradation at the start-up stage for each electrolytic cell, A predictor subnetwork configured to predict the composite cell voltage using a two-dimensional vector output by the encoder subnetwork, including, assembly.

16. The aforementioned at least one computing device: To obtain the voltage measurement value during the operation of the electrolytic device, a plurality of data acquisition and transmission units are coupled to the electrolytic cell; A processing and communication unit coupled to the aforementioned plurality of data acquisition and transmission units; A main computer server coupled to the aforementioned processing and communication unit; The assembly according to claim 15, including the assembly described in claim 15.

Images