Fault cause determination for a machine line

Abstract

The invention relates to a method for determining the cause of a fault in a machine line, in particular in a machine line for filling and packaging food and / or beverages. The machine line comprises a plurality of machines. The method is used to detect operating states for each of the machines and store same as a data set. The data set comprises the dimensions M x S x T, wherein M is the number of the plurality of machines, S is the number of different operating states, and T is the number of time steps. After an operating standstill of a machine is detected, the data set is reduced in order to determine the cause of the fault. This is carried out by determining a characteristic period of time for each of the plurality of machines. The characteristic period of time for a machine is based on a statistical cause propagation time for a fault which is caused by an adjacent machine.

Description

[0001]Determining the Cause of a Fault in a Machine Line The invention relates to a method for determining the cause of a fault in a machine line, in particular in a machine line for filling and packaging food and / or beverages, and to a machine line. Today's filling and packaging systems in the beverage and liquid food industry are highly optimized and process up to 120,000 units per hour. A typical filling and packaging system comprises a multitude of different machines and modules that are connected to one another via conveyor belts. The units, such as bottles, cans, pallets,Containers or similar items are transported from one machine to the next and pass through the individual machines on the line in a predetermined sequence. The failure of a single machine in such a machine line represents a disruption that spreads along the material flow, forcing other machines to shut down. To maintain performance at a high level, a continuous analysis process must be in place to identify and consistently eliminate weak points within the line. Therefore, in the event of a malfunction affecting multiple machines, it is important to identify as quickly as possible the machine or intermediate areas between two adjacent machines on the line that are responsible for the line or part of the line coming to a standstill. Until now, this task was primarily accomplished through specialist knowledge, with an expert in the complex machine line examining the system based on experience.to identify the error. However, this manual procedure is mostly characterized by empirical values ​​and is usually time-consuming. There is therefore a need for improved machine lines and improved methods for determining the cause of an error in a machine line. This object is achieved according to the invention by a method according to claim 1 and a machine line according to claim 9. Embodiments and further developments are covered in the subclaims. One embodiment of the invention relates to a method for determining the cause of an error in a machine line. The machine line can comprise a plurality of machines. Various materials, such as empties, bottles, cans, pallets, containers or liquids,can be transported along an arrangement between the individual machines. In a first step, an operating state of the machine can be continuously recorded for each of the plurality of machines at a specific sampling rate and stored as a data set in a memory. The data set can have the dimensions M x S x T, where M is the number of the plurality of machines, S is the number of different operating states, and T is the number of time steps. In the next step, an operational downtime of a first machine (e.g., a lead machine) in the machine line can be detected. For further, more efficient analysis of the cause of the fault, the data set can be reduced in a next step. Reducing the data set can include determining a characteristic time window for each of the plurality of machines. A characteristic time window for a first machine is based on a statistical fault propagation time for a fault,which is caused by a neighboring second machine. For each characteristic time window, an operating state ratio BV of recorded operating states within the respective characteristic time window can be recorded. The data set with the dimensions M x S x T can thus be replaced by a reduced data set comprising the dimensions M x S. Based on the reduced data set, the operating state ratios for each of the plurality of machines from the respective characteristic time window can be evaluated, and the cause of the fault responsible for the downtime of the first machine can be determined. One embodiment of the invention relates to a machine line in which the above method is implemented. Exemplary aspects of the invention are illustrated in the drawings. They show: Figure 1: a diagram showing a production line with k work areas M,each consisting of n machines; Figure 2: a diagram showing a graphical representation of the machine speed against the main flow of a filling line, as is exemplary for machines of a returnable glass line; Figure 3a: an exemplary Gantt diagram of an exemplary error propagation in a machine line; Figure 3b: an exemplary time delay between starts and stops of two machines due to a fault cause that is responsible for the downtime of a first machine; Figure 4: an exemplary histogram for the distribution of start delays of a machine after a neighboring machine is started; Figure 5: exemplary continuous probability density functions for the time delay between causally related stops and restarts of two neighboring machines; Figure 6: an exemplary causality score between two neighboring machines; Figure 7: a diagram,which shows the distribution and use of characteristic time windows; Figure 8: an exemplary flowchart for a method for determining a fault cause in a machine line; Figure 9: an exemplary system configuration for PET containers and adhesive packs; Figure 10: an exemplary system configuration for PET containers and shrink packers; Figure 11: an exemplary system configuration for cans or glass bottles; and Figure 12: an exemplary system configuration for cans. One aim of the invention is the automatic detection of the fault-causing machine for each line downtime. This should be achieved with high accuracy and low implementation costs in a filling line. To achieve good accuracy,Two types of implementation can be used: either by using a rule-based system without supervised learning or by using supervised machine learning. The embodiments described herein particularly relate to the implementation as a rule-based system without supervised learning. For this purpose, an unsupervised algorithm for extracting temporal features of error propagation in a filling and packaging line is presented. The results are usedto automatically configure rule-based algorithms for root cause analysis. First, the background for determining the causes of a plant downtime will be explained. To this end, the concept of a filling and packaging line, including its components and control mechanisms, will be presented below, and the characteristics of error propagation through a line will be explained. The available data and various approaches to data preprocessing will then be presented. Filling and packaging plants can be very diverse, ranging from processing cans to plastic bottles to glass bottles. There are also filling plants that use new bottles and those that use reusable bottles. Finally, each plant is individually tailored to the needs of the customer, their building, and their products. These points demonstrate,that a generic analysis is a challenging task. However, there is standardized data that can be collected from any machine, and there are recurring patterns from the line's perspective that can be used as a database for developing algorithms that offer great added value. Examples of different plants are described in more detail in Figures 9 to 12. The definition of a production line in general can be formulated as follows: "Production lines consist of material, work areas, and storage areas. The material flows from a work area to a storage area to a work area; it passes through each work and storage area exactly once in a fixed sequence." A generalized diagram for this type of line is shown in Figure 1 and describes the material flow through storage areas and work areas. Figure 1 shows a diagram of a production line with k work areas M,each consisting of n machines. k+1 buffers B are used to store material between work steps. The storage areas serve as buffers between production steps and allow a work area to continue production for a while, even if, for example, the previous work step does not forward any material due to an error. Packaging and filling systems represent a special case of a line and differ from the general scheme in some special features. These are described in detail below. While a production line generally has a constant production speed across all work areas, the machine speeds along the bottle flow of a filling system are not constant. This is due to the presence of a central unit, also called the lead machine. The task is toto achieve continuous operation of the lead machine. For this reason, the buffers between the machines are optimized to isolate the lead machine as much as possible from failures of the surrounding units. For example, the buffer in front of the lead machine should be as full as possible to allow the lead unit to maintain maximum continuous production in the event of a failure of the upstream unit along the bottle flow. If we continue to consider this example in the case where the fault has been rectified, the buffer naturally has less inventory than before. This quantity of missing bottles in the buffer must be replenished. Therefore, the upstream machines must be able to operate faster than the lead machine. This is referred to as overcapacity. In this example, the entire line is set up using a V-diagram.as shown by way of example in Figure 2. Figure 2 shows the machine speeds along the bottle flow of a filling and packaging line. Typically, the filling machine is the lead machine for quality reasons, for example, to avoid heating and oxidation of the product due to frequent stops. However, this is only an example, and in principle, any machine in the line can be the lead machine. Due to the very high production speeds of, for example, 60,000 bottles per hour (almost 17 bottles per second) on a glass line and up to 120,000 cans per hour on a canning line, and the associated need for high mechanical precision, the probability of unit failures or machine breakdowns is significantly higher than at slower production speeds. Therefore, the machine line must be prepared to handle most failures that occur on the line.by utilizing their buffers. Furthermore, there is always the need to find the source of frequent defects in a line in order to eliminate the root cause. Due to the interaction between machines, defects can propagate within a line. Figure 3a shows how the defect occurs as a standstill on the palletizing machine and propagates throughout the line via machines that enter the "jam" state. As this example demonstrates, the defect propagates in both spatial and temporal dimensions. The temporal progression of defect propagation is highly dependent on the occupancy of the intermediate belts and is therefore subject to significant fluctuations. Although the spatial defect propagation path usually occurs via the bottle flow, other paths besides the main flow, such as empty crate transport, are possible and occur in practice. For defect analysis, it is necessary toto have the most detailed history possible of all or all relevant machine states. Machine state describes the operating state of a machine. Various types of data are presented below that can be available in a standardized manner for most machines. However, the data are only examples, and other, less, or more data may be available. Filling machines can usually provide standardized machine states, modes, and programs according to the Weihenstephan Standard. Example machine states, i.e., operating states of a machine, are listed in the following table: Machine State Description P, roduktiv The machine produces M angel The machine is stopped due to a lack of material at input R ückstau The machine is stopped due to lack of space at exit E igenstörungThe machine is stopped due to an internal error Planned downtime The machine is stopped due to a planned task (changeover or cleaning) During operation of the machine line, an operating state of the respective machine is continuously recorded for each of the numerous machines (with a specific sampling rate). The recorded values, i.e. the recorded operating states for each machine, can be stored as a data set in a memory. The machine or operating states can be used for further processing and analysis of the fault and downtime analysis. Furthermore, the machine state data can also be used for machine learning, for example for training but also for inference. For each line downtime (e.g. stop of the lead machine), a multivariate time series of machine states can be extracted from the data, as shown as an example in Figure 3a.This results in a data set with the dimensions M x S x T, where M describes the number of machines, S the number of different operating states and T the number of time steps. A data set that describes the operating states (e.g. 10 different possible states) over a certain time period (e.g. 300 time steps) for a large number of machines (e.g. 20 machines) therefore comprises a large data volume of 20 x 10 x 300 = 60,000 data points. While it is in principle possible to use such data volumes for further processing with modern computers, the present invention makes it possible to both reduce the computing load and improve adaptability to different systems. According to one embodiment, the data set can be preprocessed. For this purpose, the time information is processed in such a way that not every time step needs to be evaluated.To achieve this, according to one embodiment, a time window is defined for each machine, which is cut from the entire time series of a plant for a specific downtime of a machine (e.g., the lead machine). The time windows are selected such that they contain all the information needed to analyze the corresponding error propagation. For illustrative purposes, Figure 3b shows error propagation from one machine to another. In this case, two machines m are connected. i and m j shown that are directly connected to each other, e.g., by a conveyor belt. The goal is now to extract the temporal coupling of the machines when one stops the other. For example, m i and thus no longer brings any bottles onto the conveyor belt. Consequently, jcontinue working only as long as there are still bottles on the conveyor belt and use the buffer function of the conveyor belt. The given example is shown in Figure 3b as a so-called Gantt chart, which graphically represents the temporal sequence of states in the form of bars on a time axis. This figure also visualizes the variables ^ and t used to describe the temporal problem. For the example described, the duration between the downtimes of m i and m j be given by: where the number of bottles on the conveyor belt connecting mi and mj is denoted as Nij, while vcur,j represents the current speed of mj. The opposite effect in the line is the restart of the machines. In this case, mi must first start production in order to provide bottles on the conveyor belt for processing to m jmust be transported. This time delay is called ^^^^^^ ^ ^ However, both delays are not necessarily constant across multiple events for many reasons. The responsible operator is always an influencing factor. For example, they often have to acknowledge a message from the machine before restarting. In addition to the operator, machines can be "prevented" from operating with another machine, which means that they have a direct signal exchange and thus, for example, one of the two machines in our example has to wait for a signal from a third machine. In addition, the state of the conveyor before the error propagation can strongly influence the time delay. In the event that the machines stop, the conveyor can serve as a buffer. Depending on the fill level (linearly dependent on the number of bottles N ij) of this buffer, the subsequent machine has to process a different number of bottles, which is directly proportional to the remaining productive time. Consequently, each measured t can be interpreted as a random event of the random variable ^. For better readability, σ will be either 'Start' or 'Stop' in the following. For example, t ^ used in formulations that apply to both the stop and start processes. To determine the probability distribution ^ ^ (t) about t ^^^^^ and t ^^^^^ for each connection (machine i to machine j) in the line, i.e., to extract each ^ ^^ ^ ^ To extract (t), for example, the following steps can be carried out: (i) Collecting the data of e.g. 10 productive days of a line and extracting the random events t^ ^ ^. (ii) Create a discrete probability distribution as shown in Figure 4. (iii) Approximate the discrete distribution by a continuous probability density function ^ ^^ ^ ^ (t). Since the stochastic processes corresponding to the random variable t exhibit stationary and independent influences, the underlying process is a Levy process. The inverse gamma function, as a member of the Levy distribution family, has empirically proven to be the best fit. An example of the exact calculation and approximation of the probability distribution is shown in the appendix. The results of the statistical processing of the measurements will be discussed below. Figure 4 shows an example histogram with the distribution of 716 events for the start delay time t. ^^^^^The events are displayed as bars, and the inverse gamma function is shown as a line fitted over the data. The maximum of the inverse gamma function can be referred to as the characteristic restart duration. Figure 5 shows continuous probability density functions for the time delay between causally related stops and restarts of two neighboring machines. ^^^^ should be larger than т due to the buffer mechanism of the conveyor belts ^^^^^ . Assuming that u ^^^^ nd ^^^ describe properties of causally related interruptions, a causality score can be calculated that indicates a probability of whether machine ^ a standstill ^ ^by machine ^. An exemplary mathematical analysis of the causality score is shown in detail in the appendix. In the following, the causality score of the appendix can be assumed. Thus, the following probabilities can be calculated: 1. a stop of m i before time t resulted in a stop of m j : 2. a restart (restart) of mi after the time t leads to a restart (restart) of mj: т^^^^^^ ^^^^^ ^ ^ < ^ ^ − ^ This results in a time-dependent causality score, which can be used to describe the probability of whether the failure of one machine is or could be responsible for the failure of the neighboring machine. A possible example of the causality score is shown in Figure 6. As can be seen in Figure 6, for two neighboring machines, m i and m ja relatively clear time period can be identified in which a machine failure can occur i causally the cause of the failure of the neighboring machine is responsible. For example, if the machine m i and then the machine m j the probability is relatively high that the neighboring machine m i for machine failure is responsible. However, if one of the machines fails 400 seconds after the other, the causal probability is low. The consideration, or rather the calculation and use of the causality score is not restricted to two neighboring, i.e. directly connected, machines. By mathematically convolving the various probability density functions of the corresponding machines and machine connections, a score can be obtained for the consideration of any two machines which describes the causality over a large area. From these considerations, certain time windows arise for the failure of a machine in a machine line consisting of many machines, which time windows are particularly interesting for fault analysis. As described above, recording all operating states results in very large amounts of data.During operation of a machine line, an operating state of the machine is continuously recorded for each of the numerous machines at a specific sampling rate. For further processing, the data can be stored in a two-dimensional or three-dimensional data structure. The data structure can include the dimension of the number of possible states. The resulting data set therefore has the dimensions M x S x T, where M is the number of the numerous machines, S is the number of different operating states, and T is the number of time steps. In order to reduce this large data set for further processing, the knowledge of the relevant time windows can be exploited. For this purpose, the temporal behavior of error propagation can be used as described above.Based on the extracted probability density functions, characteristic runtimes between start / stop of two connected machines m ^^^^^ ^^^^ i and mj can be calculated as ^. ^^ and ^^^. For simplicity, the running times are referred to below as the characteristic stopping delay time т ^^^^ and characteristic start-up delay time т ^^^^^These characteristic values ​​can be automatically determined based on the above findings derived from the fitted density functions. For example, based on causality information as shown in Figure 6, a characteristic time window can be determined for each of the machines in the machine line, and only the state information from the corresponding time windows can be considered. Thus, after detecting a downtime of a first machine (e.g., a lead machine), a characteristic time window can be determined for each of the plurality of machines. As described above, a characteristic time window for a first machine is based on a statistical fault propagation time for a fault caused by a neighboring second machine.In an example implementation, the time interval in which the probability of a fault being caused by a neighboring machine is greater than a threshold value can be selected for a characteristic time window for a machine. The operating state during each characteristic time window can then be evaluated. There are various ways to do this. For example, an operating state ratio BV of recorded operating states within the respective characteristic time window can simply be determined. Or the operating states can simply be used as samples in the time window without further processing. The original data set with dimensions M x S x T can thus be replaced by a reduced data set. The reduced data set comprises dimensions M x S.An example of setting characteristic time windows is shown in Figure 7. A characteristic time window for a first machine is based on a characteristic stopping delay time т. ^^^^ and a characteristic start-up delay time т ^^^^^ . The characteristic stopping delay time т ^^^^ of a first machine can be defined as a period of time from a starting point of a standstill of a directly connected neighboring machine to the starting point of a standstill of the first machine. During the characteristic stop delay time t ^^^^ the first machine continues to run normally. The characteristic start-up delay time t ^^^^^ of the first machine can be a time period from a starting point of a start-up of the directly connected neighboring machine to the starting point of a start-up of the first machine. During the characteristic start-up delay time ^ ^^^^^the first machine is still at a standstill. The respective characteristic time windows for the plurality of machines are based on the arrangement and connection of the plurality of machines in the machine line. Based on the reduced data set, the operating status conditions or the operating states for each of the plurality of machines from the respective characteristic time window can be evaluated, and the cause of the fault responsible for the downtime of the first machine can be determined. For the evaluation and fault analysis, rule-based algorithms, a classification method, and / or a relational graph convolutional network, for example, can be used. Each of these possible methods benefits from the data preprocessing described above and the reduced data set obtained from it.A further advantage of the data preprocessing described above is that the temporal information for error propagation between two machines can be used for each different arrangement of machines in a machine line. Thus, the same rule-based algorithm, the same classification method, and / or the same relational graph convolutional network can be used for different machine arrangements. For implementation, only the characteristic time window for each machine needs to be adjusted according to the different arrangement of machines in each new machine line. As mentioned above, evaluating the operating state conditions and determining the cause of the error can be performed using a rule-based algorithm that successively evaluates the characteristic time windows.The rule-based algorithm preferably begins with the characteristic time window of the stationary master machine and continues with a characteristic time window of a neighboring, connected machine. Alternatively, the evaluation of the operating state conditions and determination of the cause of the fault can be carried out using a classification method or a relational graph convolutional network. The classification method or the relational graph convolutional network can then be trained such that it can calculate a cause of the fault based on reduced data sets of dimensions M x S. Figure 8 shows a flowchart of an exemplary method for determining a cause of a fault in a machine line. In step S802, an operating state of the machine is continuously recorded for each of the plurality of machines at a specific sampling rate and stored as a data set in a memory.The data set has the dimensions M x S x T, where M is the number of the plurality of machines, S is the number of different operating states, and T is the number of time steps. In step S804, an operational downtime of a first machine (e.g., a lead machine) of the machine line is detected. For further more efficient analysis of the cause of the fault, the data set is reduced in step S806. Reducing the data set comprises steps S806a-c, as described below. In step S806a, a characteristic time window is determined for each of the plurality of machines. A characteristic time window for a first machine is based on a statistical fault propagation time for a fault caused by a neighboring second machine. In step S806b, an operating state ratio BV of detected operating states within the respective characteristic time window is recorded for each characteristic time window.In step S806c, the data set with the dimensions M x S x T is replaced by a reduced data set comprising the dimensions M x S. Based on the reduced data set, in step S808 the operating state conditions for each of the plurality of machines from the respective characteristic time window are evaluated and the cause of the error responsible for the downtime of the machine in question is determined. The following Figures 9 to 12 describe various exemplary system configurations for various bottle filling systems in which the invention or at least parts and aspects of the invention can be implemented. The description of Figures 9 to 12 is intended only to provide a general overview of machines in which errors can propagate and in which machine downtimes can occur. Figure 9 shows an exemplary system configuration 1000 for PET bottles or PET containers and adhesive packs.As can be seen in Figure 9, the system configuration 1000 comprises a variety of modules that form a line at the end of which finished filled PET containers are output in the form of a bundle on pallets. Some of the modules and machines can be optional, and the invention is not limited to the precise form and arrangement of the system configurations. The system configuration 1000 comprises an oven 1002 for preforms, a preform sorter with feeding machine 1004, and a blow molding machine 1008. The modules 1002, 1004, and 1008 generally form a stretch blow molding machine in which PET containers are produced and formed from a starting material. The produced PET containers are passed on to a filler 1010, in which the bottles are filled. The filler can optionally include a rinser. Various particles such as dust, cardboard, or residue from wooden pallets can settle in the preforms during storage or transport.These can be removed using the rinser. A closer can be arranged at the end of the filler, by means of which the PET containers are closed after filling. Optionally, the system configuration 1000 can include a rotating device downstream of the filler 1010, which is used for hot filling of the PET containers. Via one or more conveyor belts 1016, which can also include a buffer 1018 for temporarily loading filled containers, the filled PET containers are guided to a separator 1020 and then to a drying device 1024, in which the PET containers are dried. After drying, the PET containers are conveyed to a labeling machine 1026. The labeling machine 1026 can be designed for various labeling techniques, such as labeling using hot melt, cold melt, self-adhesive labels, or sleeves.After the PET containers have been printed or labeled, they are passed through a second drying device 1028, a line distributor 1030, conveyor belts 1032, an adhesive pack production line 1034, and a curing section to a handle applicator. In the adhesive pack production line 1034, the PET containers are grouped into specific group sizes and packaged into a pack, such as a "six-pack." In the handle applicator, a carrying handle is attached to the pack, allowing for comfortable carrying. The finished packs are then arranged accordingly by a robot 1042 for layer production and packed on pallets by a palletizer 1044. In the system configuration 1000, so-called format trolleys or format racks can be arranged on various modules and machines to provide quickly interchangeable format sets for short changeover times and automatic tool changes.Examples of format carriages are the format carriage 1006 for the blow molding machine 1008, the format carriage 1012 for the filler 1010, the format carriage 1022 for the labeling machine 1026, the format carriage 1038 for the adhesive pack production 1034, and the format carriage 1046 for the palletizer 1044. Figure 10 shows another example system configuration 1100 for PET containers and shrink packers. The system 1100 in Figure 10 includes many of the modules and machines from the system configuration 1000 in Figure 3, but there are some differences. The description of the modules already described in connection with Figure 9 is therefore omitted for Figure 10. A key difference between the two exemplary system configurations 1000 and 1100 is that the labeling machine 1126 with the labeling modules 1127 can already be installed after the blow molding machine 1008 and before the filler 1008.For this purpose, the system configuration 1100 can comprise six transport lanes 1150 into which the PET containers can be pushed. After the PET containers have pushed into one of the six lanes 1150, they are conveyed into the film wrapping module 1152 and then into the shrink tunnel 1154. Figure 11 shows an exemplary system configuration 1200 for cans or glass bottles. The exemplary system configuration 1200 from Figure 11 again has some similarities to the system configurations 1000 and 1100 from Figures 9 and 10 and the description of the system configuration is therefore limited to the differences in the system configurations. As shown in Figure 11, the exemplary system configuration can comprise two separate feeds. A first feed, on the left in Figure 11, shows a branch for cans or optionally a partial branch for reusable new bottles. The containers, ieCans or new bottles are fed into the machine by a depalletizer 1302, where they are guided via conveyor belts to the filler 1010. A second feed, on the right in Figure 11, shows a sub-branch for reusable bottles, which are fed into the system from a reusable sorting system (not shown). If the already used reusable bottles are fed into the system 1200 via the sub-branch for reusable bottles, the reusable bottles first pass through the cleaning machine or washing machine 1304. Another possible difference in the exemplary system configuration 1200 is the transfer packer 1306 after the labeling machine 1026. The transfer packer can sort the bottles or cans into a carton clip application or into crates, or both. Figure 12 shows an exemplary system configuration 1300 for cans, in which the elements already described in the other system configurations are again omitted.The cans in system configuration 1300 are fed into the depalletizer 1302 from a magazine 1402 containing cans. After passing through the filler and being filled, the cans are closed using a closure magazine 1404 and transported further along the system 1400 via the conveyor belts, as described above. The optional pasteurizer 1408 can be bypassed via the bypass 1412 if not required. In the pasteurizer 1408, the freshly filled products can be pasteurized for preservation. In contrast to the system configurations 1000, 1100, and 1200, the example system configuration 1300 shows various tanks for corresponding consumables, such as tanks 1410 with rinsing liquid and / or the filling product and tanks 1406 with belt lubricant. These tanks can also be included in the example system configurations already described above.For example, the tanks 1406 and 1410 can store the chemical products 106 that are fed from the mixer 110 to the machines. Appendix Gamma Function The inverse gamma distribution is defined as follows: with corresponding cumulative distribution function where Γ(α, x) is the upper incomplete gamma function: Γ(α) is the gamma function: and Q is the regularized gamma function By approximating the given discrete density function by the inverse gamma function, one obtains as shown in Figure 5 and can ^ ^^ ^ ^ as the probability for a time delay between causally related stops and restarts of two neighboring machines: For better readability, the following is used: Causality between standstills Assuming that ^^^^^^ and ^^^^^^ ^ ^ ^^To describe properties of causally related standstills, a causality value or (causality score) can be calculated i j ( ) : R + (3.10) which describes whether the machine i reaches a certain standstill S v of machine j. ^ describes the space of all possible time series Ti of machine states on machine i and ^ the space of downtimes containing start and end time information. For each t ∈ T i the probability ^ is calculated that 1. a stop of mi before time t led to a stop / standstill of mj: ^ ^^^^ ^^^^ ^^ > ^^ − ^ 2. a restart (restart) of m i after the time t to restart (restart) of m j leads: Compare to Figure 3b for an illustration of the temporal constraints. The probability of an error in mi at time t is equal to the probability that both of the listed dependencies are satisfied. Therefore, As shown in Figure 6. The probability s can also be given by ^^ ^ ^ be expressed as follows: If one integrates sij(t, S) over the duration of a potentially causative standstill Sc at mi, the causality score c for this standstill (Sc) is as follows This score has the potentially causative interval Sc and the victim interval Sv as input. By summing sij(t, S) over all periods in which mi is in a self-inflicted error ℰ ∈ ^, the causality score for m can be calculated. i as the cause of the stop S at m j determine. Further causality The result of the previous section (equation 3.15) is derived. However, so far, fitting has been done on data that should be collected under the assumption of causality between the stop / start events and can therefore only be extracted for direct neighbors. However, under this restriction, the causality of stops can only be calculated for directly connected machines. To circumvent this restriction and also capture higher-order connections, ^ by folding all ^ ^^^^ with |i′ - j′| = 1 along the direct connection between machines i and j: Using the convolution formula the distributions for calculating the causality score for, for example, two machines i and j selected under the condition j > i can be expressed as follows

Claims

Patent claims 1. Method for determining a cause of a fault in a machine line, in particular in a machine line for filling and packaging food and / or beverages, wherein the machine line comprises a plurality of machines, wherein material is transported along an arrangement between the individual machines, and wherein the method comprises: for each of the plurality of machines, continuously recording, at a specific sampling rate, an operating state of the machine and storing the recorded operating states as a data set in a memory, wherein the data set has the dimensions M x S x T, where M is the number of the plurality of machines, S is the number of different operating states, and T is the number of time steps; detecting an operational standstill of a first machine in the machine line;Reducing the data set, comprising: determining a characteristic time window for each of the plurality of machines, wherein a characteristic time window for a machine is based on a statistical fault propagation time for fault propagation between two neighboring machines; determining, for each characteristic time window, an operating state ratio BV of detected operating states within the respective characteristic time window; replacing the data set having the dimensions M x S x T with a reduced data set comprising the dimensions M x S; and based on the reduced data set, evaluating the operating state ratios for each of the plurality of machines from the respective characteristic time window and determining the fault cause responsible for the downtime of the first machine.

2. The method of claim 1, wherein:; The evaluation of the operating state conditions and determination of the cause of the fault is carried out using a rule-based algorithm, a classification method, and / or a relational graph convolutional network. The same rule-based algorithm, the same classification method, and / or the same relational graph convolutional network is used for different arrangements of machines in a second machine line. The characteristic time window for each of the machines is adapted according to the different arrangement of machines in the second machine line. 3.The method according to claim 1 or 2, wherein the evaluation of the operating state conditions and determination of the fault cause are performed by a rule-based algorithm that successively evaluates the characteristic time windows, wherein the rule-based algorithm begins with the characteristic time window of the stationary first machine and continues with a characteristic time window of a neighboring connected machine.

4. The method according to claim 1 or 2, wherein the evaluation of the operating state conditions and determination of the fault cause are performed by a classification method or by a relational graph convolutional network, wherein the classification method or the relational graph convolutional network has been trained such that it can calculate a fault cause based on reduced data sets of dimensions M x S. 5.The method according to claim 4, wherein the data set with dimension M is processed into the same number M of classes using a neural network as part of a classification task.

6. The method according to one of claims 1 to 5, wherein a characteristic time window for a first machine is based on a characteristic stopping delay time t. ^^^^ and a characteristic start-up delay time т ^^^^^ where: the characteristic stopping delay time т ^^^^ of a first machine is a time period from a starting point of a standstill of a directly connected neighboring machine to the starting point of a standstill of the first machine, during the characteristic stop delay time т ^^^^ the first machine is still running properly; and the characteristic start-up delay time т ^^^^^of the first machine is a time period from a starting point of a start-up of the directly connected neighboring machine to the starting point of a start-up of the first machine, during the characteristic start-up delay time т ^^^^^the first machine is still at a standstill.

7. The method according to claim 6, wherein the respective characteristic time windows for the plurality of machines are based on the arrangement and connection of the plurality of machines in the machine line.

8. The method according to one of claims 1 to 7, wherein the machine line further comprises regions between the plurality of machines, and a cause of a fault in an intermediate region between two adjacent machines can be identified by evaluating characteristic time windows.

9. A machine line, in particular in a machine line for filling and packaging food and / or beverages, wherein the machine line comprises: a plurality of machines connected to one another in an arrangement, material being transported along the arrangement between the individual machines;a sensor for each of the plurality of machines, each configured to continuously detect, at a specific sampling rate, an operating state of the machine and create a data set, wherein the data set has dimensions M x S x T, where M is the number of the plurality of machines, S is the number of different operating states, and T is the number of time steps; a memory for storing the data set; and a computing device configured to: detect an operational downtime of a first machine of the machine line, reduce the data set, comprising: determining a characteristic time window for each of the plurality of machines, wherein a characteristic time window for a machine is based on a statistical error propagation time for an error caused by a neighboring machine; Determining, for each characteristic time window, an operating state ratio BV of recorded operating states within the respective characteristic time window, and replacing the data set with dimensions M x S x T with a reduced data set comprising dimensions M x S, and based on the reduced data set, evaluating the operating state ratios for each of the plurality of machines from the respective characteristic time window and determining the cause of the fault responsible for the downtime of the first machine. 10.The machine line according to claim 9, wherein: the evaluation of the operating state conditions and determination of the fault cause is performed by a rule-based algorithm, a classification method, and / or a relational graph convolutional network; the same rule-based algorithm, the same classification and regression method, and / or the same relational graph convolutional network is used for different arrangements of machines in a second machine line; and the characteristic time window for each of the machines is adapted according to the different arrangement of machines in the second machine line. 11.Machine line according to claim 9 or 10, wherein the evaluation of the operating state conditions and determination of the cause of the fault is carried out by: a rule-based algorithm which successively evaluates the characteristic time windows, wherein the rule-based algorithm begins with the characteristic time window of the stationary first machine and continues with a characteristic time window of a neighboring connected machine; or a classification method or is carried out by a relational graph convolutional network, wherein the classification method or the relational graph convolutional network has been trained such that it can calculate a cause of the fault based on reduced data sets of the dimensions M x S, wherein the data set. with the dimension M to the same number M of classes is processed by means of a neural network as part of a classification task; or wherein an area between two machines is identified as the cause of the fault that is responsible for the downtime of the first machine.

12. Machine line according to one of claims 9 to 11, wherein a characteristic time window for a first machine is based on a characteristic stop delay time t ^^^^ and a characteristic start-up delay time т ^^^^^ where: the characteristic stopping delay time т ^^^^ of a first machine is a time period from a starting point of a standstill of a directly connected neighboring machine to the starting point of a standstill of the first machine, during the characteristic stop delay time т ^^^^the first machine continues to run normally; and the characteristic start-up delay time т ^^^^^ of the first machine is a time period from a starting point of a start-up of the directly connected neighboring machine to the starting point of a start-up of the first machine, during the characteristic start-up delay time т ^^^^^ the first machine is still standing still.

Images