EP3774267A1

EP3774267A1 - Method for the automatic process monitoring and process diagnosis of a piece-based process (batch production), in particular an injection-molding process, and machine that performs the process or set of machines that performs the process

Info

Publication number: EP3774267A1
Application number: EP19715418.0A
Authority: EP
Inventors: Stefan Kruppa; Stefan Moser; Matthias BUSL; Reinhard Schiffers
Original assignee: KraussMaffei Technologies GmbH
Current assignee: KraussMaffei Technologies GmbH
Priority date: 2018-03-27
Filing date: 2019-03-26
Publication date: 2021-02-17
Also published as: CN111867806A; US20210008774A1; WO2019185594A1; KR20200131302A; MX2020009671A; DE102018107233A1

Abstract

The invention relates to a method for the automatic process monitoring and/or process diagnosis of a piece-based process, in particular a production process, in particular an injection-molding process, comprising the steps: a) performing an automated reference finding in order to obtain reference values (r ₁... r _n) from values (x ₀... x<sb /> _j) of at least one process variable; b) performing an anomaly detection on the basis of the reference values (r ₁... r _n) found in step (a); • c) performing an automated cause analysis and/or an automated fault diagnosis on the basis of a qualitative model of process relationships and/or on the basis of dependencies of various process variables on each other.

Description

description

Method for automatic process monitoring and process diagnosis of a piece-based process (batch production), in particular an injection molding process and a machine performing the process or a machine park performing the process

The invention relates to a method for automatic process monitoring and for the diagnosis of a piece-based process and a machine performing the process Ma, in particular an injection molding machine, or a process implementing machinery.

Process monitoring and / or process diagnostics are generally based on fixed limits that must first be set manually. This means that a process variable or a characteristic has an upper and lower limit which, for example, has to be determined based on experience of the operating personnel and in particular must be set manually in the control or in an operating data acquisition system. Furthermore, it is known to make a threshold value over several levels recognizable, for example by means of a warning upstream warning.

On the basis of this, the stability of the process or the process capability, ie the ability of the process to be evaluated, and in the case of leaving the intended process, eg. B. when exceeding or falling below the upper or lower limit measures are initiated, the z. B. may include reject sorting and alerting.

From the technical publication of 07/14/2015 entitled Priamus FILL CONTROL 1.13 Release Notes of Priamus System Technologies AG, Schaffhausen, Switzerland is a concept for process monitoring by means of a so-called Q-Button Priamus System Technologies AG in 8200 Schaffhausen, Switzerland. With this technology, the limit value determination of the upper and / or lower limit value is performed semi-automatically by means of a so-called Q-button. This is an operating device that automatically sets limit values on the basis of "six-sigma values", which ensure a sensible setting for the monitoring in an optimized process.

Furthermore, it is from the Western Electric Company (1956), Statistical Quality Control Handbook; 1.ed., Indianapolis, Indiana: Western Electric Co. is known to determine from a reference the standard deviation and to generate therefrom with fixed rules alarms concerning parameters or controlled variables.

Common to all prior art approaches is that they generate singular alarms or other actions for each tolerance violation of a particular parameter, without taking into account a possible interaction of the individual parameters / limit value overruns. In other words, there is no causal connection between the limit violations taken, so that any existing disturbance, the z. B. can affect different values, can not be reliably detected.

Machine learning methods are capable of automatically detecting anomalies and even diagnosing them. However, these first require data that reflects the corresponding faults and the associated causes. Therefore, they are only able to provide known or already occurred diagnoses and repeat if necessary. In addition, it is difficult to create universally valid models through these methods because they can not distinguish between specific and general contexts.

Such methods are already used to predict quality (see US 7,216,005 B2). In such methods, however, the algorithm must first be trained specifically for a process. The proposed methods are therefore not self-learning and executable. Furthermore, expert systems and diagnostic methods based on qualitative models are known, for example, by the term "Model-Based Diagnosis" (compare R. Reiter, A theory of diagnosis from first principles, Artificial Intelligence 32 (1) (1987) 57-95). ,

The prior art outlined above has a number of disadvantages. Due to the manual definition of the limits, two conditions must be met for effective monitoring:

1. A limit must be set and

2. The monitoring must be applied.

Limit values can be determined via tests and / or automatically derived from them. Nevertheless, the trial period and / or the data for this must be communicated ex clicitly to the program which assumes the machine control / process control.

The following is explained using the example of an injection molding process. With the approximately 100 process variables of a modern injection molding machine (see actual value cycles), in practice only limits are set for the fewest process variables. Likewise, the monitoring options integrated in the machine and in external systems (MES) are not always used because the limit values may also have to change depending on the machine used, environmental factors and material / material properties for a process to achieve consistent quality control.

As a result of the resulting high workload required for holding the limit values, a large number of process variables of the abovementioned approximately 100 process variables usually remain unsupervised in practice.

Only the most important functions are updated to manually enter limits adapted to current environmental conditions. Thus, the control potential, in particular as far as a theoretically possible limit value monitoring is concerned, remains unused in many areas, since a full use of the potential means a very high updating effort and support effort by the serving staff.

A further disadvantage is that the information about which tolerance violations occur or in which way the tolerances are exceeded (for example, once, permanently, creeping and / or becoming increasingly strong and / or waning, etc.), does not mean any further automatic Conclusions can be drawn from this information. Thus, it can be quite possible that multiple tolerance transgressions occur simultaneously, which have a common unambiguous cause, without the se being named, recognized and thus analyzed purposefully.

Rather, it is the experience of the operator to recognize a possible common cause and to remedy this with a certain characteristic combination of tolerance overruns of individual values of process variables on the basis of his specialist knowledge.

The object of the invention is therefore to avoid the above-mentioned disadvantages of the prior art and / or at least mitigate. In particular, a fully automatic process monitoring and process diagnosis, in particular for a piece-based process, which may be an injection molding process specified who the who, the method should be able to set automatically and in particular self-learning reference values and limits for process variables in order To recognize or at least even eliminate any exceedances and anomaly assessments, and to draw conclusions on any new references or limit values that may be useful.

This object is achieved by a method having the features of claim 1. Advantageous embodiments are specified in the subclaims.

An inventive method for automated process monitoring and / or process diagnosis of a piece-based process, in particular a production process in particular of an identical injection molding process, comprises the steps of: a) carrying out an automated reference finding to obtain reference values n ... r _n from values x ₀ ... X _{j of} at least one process variable;

b) performing anomaly detection on the basis of the reference values ri ... r _{n found} in step (a);

c) performing an automated root cause analysis and / or an automated fault diagnosis based on a qualitative model of the process and / or on the basis of dependencies of different process variables on each other.

With the method according to the invention, it is possible, despite a variety of possible anomalies, to sort these anomalies and take in a take for the operator, clear presentation, so that the operator due to egg ner variety of anomalies a preferably unique cause of the Hand gets, on the basis of which he has a disturbing cause, so z. B. can fix a Prozessstörgröße or otherwise disrupt the process.

Furthermore, the operator is relieved of manually setting limit values for different process variables, even if environmental conditions or the like possibly change. This can handle the inventive method automatically table and thus for a further automated process improvement and thus an increase in quality of the pieces produced, for. As injection molded parts provide.

Because corresponding limits are automatically available for a large number of process variables, the method according to the invention can also accomplish automated monitoring of all process variables and, by means of a large number of these monitored process variables, automatically provide an improved root cause analysis and cause indication.

In a preferred embodiment of the method according to the invention, a result of the root cause analysis and / or the fault diagnosis is output at an output device. output for an operator or a result of the cause analysis / fault diagnosis is processed further automatically. This can be z. B. ge happen that the result of the root cause analysis of a machine control and / or a machine park control and / or a control system for influencing a Ma environment, z. B. a workshop, z. B. whose heating / air conditioning or the same is provided. This makes it possible either to make it particularly obvious to the operator what are the causes of a certain anomaly or it can succeed in achieving an automated averting of these anomalies, for example if a machine control or an engine hall control or a machine park control according to results of Ursa chenanalyse react.

In a preferred embodiment of the method according to the invention, step a) may comprise at least one or more of the sub-steps listed below: a1) Evaluation of process values x ₀ ... X _j of process variables over several process cycles with respect to their suitability for use as reference by Calculation of evaluation parameters b... B and application of defined rules, where the change criteria of the values x ₀ ... X _{j of} the process variables, and / or fluctuations of the process variables are used as evaluation parameters bi ... bi, for example, or a2) as a reference be of the automatic process control and / or auto matic process diagnostics automatically applies certain reference values n ... r _n ver which z. B. reflect the "natural" noise or uncertainty of the process variable that each process variable has due to environmental conditions and / or sensor noise, or

aß) if the reference values r ^* i ... r ^* _n formed from the process values x ₀ ... X _j of process variables are better than the currently best found reference values ri... r _n on the basis of criteria and rules, these are set as new best found reference values ri ... r _n or a4) the reference values ri. , , r _n from step a3) are used to automatically z. As jumps, gradients, outliers can be recognized as anomalies, evaluate and / or optionally mark or

a5) whereby the automatic reference, that is to say the reference values x 1... r _n , is necessarily newly formed in the case of predetermined events, whereby such predetermined events may be, for example, a prolonged standstill of the machine carrying out the process or a tool change.

In a further preferred embodiment of the method according to the invention, each process variable is assigned a reference generator for forming reference values ri ... r _n , which is preferably provided by the manufacturer with an initial reference, from which then the development of further future references, ie reference values n. .. r _n can take place. The initial reference represents a first current reference with the reference values ri ... r _n , which can be modified with the method according to the invention, in particular in step a).

In a further preferred embodiment of the method according to the invention, a reference consists of a plurality of values ri ... r _n , wherein the values ri. , , r _n intrinsic properties of a gradient values of values x ₀ ... X _j reflect a process variable, z. The standard deviation or the median of the values.

A further embodiment of the method according to the invention is characterized in that, during the course of the process, the reference values are adjusted to the values of the values x ₀ ... X _{j of} the process variable determined by measurement, a window of j values.

In a further embodiment, provisional reference values ri ^* ... _N ^* and evaluation numbers b... B are formed from the j values of the process variable, wherein the evaluation numbers b. B. the slope or the curvature of the j values and / or the course of values over time can be.

Furthermore, it may be advantageous, from the evaluation numbers bi ... b, the current reference (value x ^... R _n) and the temporary reference (value n ^* ... r _n ^*) by means of pre- certain rules to determine whether the current reference r ^. , , r _n is maintained or provisional reference ri ^* ... r _n ^* in the future as a new current reference ri ... r _n is used and thus the provisional reference ri ^* r _n ... ^* the existing current reference ri ... r _n replaced.

It is expedient that anomaly detection is provided for each process variable, which current reference values. , , r _n and / or past values Xi ... X _j of the process variable used to an exceptional value, that is, detect an abnormality tight or too assigns an exceptional value of an abnormality probability.

Furthermore, it is preferable to have a value which z. B. is more than three reference standard deviations away from the reference mean, to mark as anomaly or evaluate e.g. by indicating the deviations from the reference mean value in multiples of the reference standard deviation. This embodiment is not limited to three times the reference standard deviation alone. Depending on the value considered, d. H. Depending on the process variable considered, a suitable deviation from the reference mean value can be defined. If appropriate, this can also be done by way of tests.

Furthermore, it is advantageous that a qualitative model of an injection molding process is used as the qualitative model used in step c), in which relationships between the process variables and / or dependencies between the process variables z. In the form of rules, e.g. B. a set of rules are included.

Such a set of rules or an accumulation of rules allows a reliable cause analysis and thus the issue of the lowest possible number of possible causes for the operator, even if, for example, a large number of anomalies has been identified.

Another object of the invention is to provide a machine, in particular an injection molding machine, with which the inventive method for automatic process monitoring and / or process diagnosis can be performed. This object is achieved with a machine according to claim 13, which is set up and designed to carry out the method according to the invention. Such a machine is in particular designed as an injection molding machine.

It is also an object of the invention to provide a machine park, in particular a Ma machine park having injection molding machines, with which the erfindungsge Permitted method for automatic process monitoring and / or process diagnostics can be performed.

This object is achieved with a machine park having the features of claim 14. Such a machine park is set up and designed to carry out / implement the method according to the Invention.

In the following the invention will be explained by way of example with reference to the drawing. Show it:

FIG. 1 shows a schematic structural diagram of an anomaly detection for a specific characteristic figure by the method according to the invention;

FIG. 2 shows a reference update after a value jump, obtained by a method according to the invention;

Figure 3: exemplary relationships that may have an effect on a process index, in particular the example of a Kunststoffspritzgießpro process.

FIG. 4 shows a flow chart for determining a new reference in a reference generator used according to the invention;

FIG. 5: a flowchart relating to an anomaly assessment. In FIG. 1, an anomaly recognition, in particular a self-referencing anomaly recognition according to step b) of the method according to the invention, is shown in a highly schematized form in the form of a structure diagram. By way of example, this is represented on the basis of a process variable (characteristic number 1) representative of any data source, in particular for process parameters or process parameters or their measured values. Such a data source (characteristic number 1) supplies values x ₀ ... X _{j of} the process variable and is supplied to a reference generator and to anomaly detection. The reference generator contains a so-called current reference with current reference values n ... r _n and thus the anomaly detection is capable of an exceptional value by comparing the process variable (measure 1) with the current reference values ri ... r _n and / or with previous values xi ... x _{k of} the process variable. For example, it is determined that a current value x _{0 of} the process variable (measure 1) is marked or evaluated as an anomaly if more than three reference standard deviations between the value to be assessed x _{0 of} the process variable (measure 1) and the reference average lie. The reference mean can be z. B. Mean value is the part of the current reference values n ... r _n and / or calculated from the previous values Xi ... X _{j of} the process variable. This may preferably be an arithmetic mean. To obtain a reference, that is, from current reference values ri ... r _n to replace a current reference by a future reference, reference is made to the description of Figure 4 below, based on the operation of a reference generator will be explained.

The reference generator is preferably present for each process variable (code) that is to be subjected to anomaly detection. The reference generator is provided for example by the manufacturer of the process performing machine with an initial reference, the first reference values n ... r _n for the value x _{0 of} the process variable (code 1) represents. Such a reference can consist of several values ri... _N , where, for example, n = 10. The reference can be z. B. a standard deviation and / or the mean and / or the like of a value curve of the process variable, ie the key figure 1. The values x ₀ ... X _{j of} the current process are read into the reference generator, whereby the reference to the process variable course is adjusted. The process variable course is a temporal course of the measured values concerning the process variable / the key figure 1. To adjust the reference here is a window of z. For example, consider j values where j is 10, for example. However, j can also easily assume values between 2 and 50 or 100, depending on exactly how a determination is to be made.

From these values j are provisional reference values ri ^* r _n ^* ... formed and evaluation numbers bi ... bi. The evaluation numbers bi ... b, z. As for evaluating the quality or the suitability of the provisional reference ri ^* ... r _n ^* to evaluate the Anomalieerken planning.

The evaluation numbers bi ... b, for example, a department, z. As a slope or a curvature or other parameter of the sequence of the corre sponding j values. From the evaluation numbers bi ... bi, the current reference n ... r _n and the provisional reference ri ^* ... r _n ^* is determined on the basis of predetermined rules and determines whether the current reference h ... r _n is maintained or whether the environmental conditions have changed in game example, that the preliminary Refe rence ri ... _rn ^* the current reference ... r _n is replaced and worked in the following with the previous provisional, now current reference ^* n ( n ^* -> n ... r _n ^* -> r _n ).

As an example, it should be noted that if z. For example, if the slope over j = 10 values is less than one standard deviation from the current reference n ... r _n and the provisional standard deviation is not greater than twice the current current reference ri ... r _n , the provisional one Reference ri ^* ... r _n ^{* is} adopted. If this is not the case, the provisional reference ri is ^* ... r _n ^* discarded and the steps of Wertesam melns and comparing start over. The process will continue until then with the unchanged current reference n ... r _n .

Such a fixed, current reference ri ... r _n is combined with the previous values x- _| ... x _{k is passed to} the process variable of the anomaly detection to determine an exceptional value x _a . Here, k is the solid of values which are considered for the anomaly detection, where k is eg 20. If a value x _a z. For example, if more than 3 reference standard deviations are removed from the reference mean, it will be marked as anomaly. However, instead of or in addition to the above-mentioned anodic Malieerkennung, in which there is the status anomaly (yes) or anomaly (no), are also fed to an anomaly likelihood determination. A certain extraordinary value x _a may, depending on the deviation from the corresponding current reference n... R _{n, have} a certain probability of anomaly, eg. B. 70 or 75% are net zugeord. Such anomaly tagging is then passed to a root cause analysis. Such anomaly detection on the basis of values of different process variables is carried out in parallel to other process variables analogously to the anomaly detection explained above. The results of the anomaly detection are transferred to the root cause analysis.

With this procedure, it is possible that a large number of anomalous messages / anomaly labels / anomaly probabilities arise at the same time or in short intervals and thus a large number of messages / warnings / alarms are generated, since a large number of process variables (characteristic numbers 1) are processed in parallel and a process problem is usually not only reflected in a single process variable, ie in a single measure.

Such a large number of messages / warnings / alarms is then channeled according to the invention by a root cause analysis and processed easily understandable for a user / operator or weitergege ben to automated systems (controllers). The cause analysis is formed as a so-called user-oriented summary of the anomalous messages and also the stability messages explained below and, as such, essential to the invention. The operator / user or the process manager is usually only interested in the causes of the process variable change, not so much in the individual process variable change as such.

The root cause analysis is carried out as a third step of the method according to the invention based on the knowledge of the relationships between the process variables, which is present in a specific process. This knowledge is often in the experience treasure of manufacturers of such machines or operators available and is once made available in the form of a data loading of the root cause analysis availa tion and stored there. The root cause analysis uses this knowledge about the Relationships between the process variables in order to conclude on causes or to make a targeted diagnosis or to make diagnostic recommendations.

For this purpose, a qualitative model of the process, in particular an injection molding process is used, which contains relationships among the process variables. Comprehensive empirical values are available in the specialist circles for this purpose. These must be filed in the source analysis, especially in the form of "if-then relationships".

So it may be z. As in an injection molding process, that an increased cylinder temperature leads to a more liquid plastic melt in Plastifizierungszylinder and thus leads to a lower pressure level during the injection process or at druckreggel tem injection to a higher injection speed. Thus, a lot of detected number of anomalies detected for individual values, for example, too low a plastic melt in Plastifizierungszylinder, too low a pressure level in the one injection process or too high an injection speed by the Anomalieerken voltage detection, based on corresponding empirical values Ursa chenanalyse a single cause can determine, namely the one that all three of the water consequences, for example, attributed to an increased cylinder temperature who can. Such a set of rules can include a great many rules and is essentially dependent on the process to be assessed or automatically analyzed. Such a multi-rule set of rules accomplishes it in accordance with the invention that on the basis of the diagnosis can be severely limited and despite a variety of detected anomalies the user / operator concrete, matching these abnormalities diagnostic result is delivered, which is a targeted intervention in the process allows. Thus, the user gets so only the Diagnosemel tion, which is interesting for him and can thus faster identify the cause of the changes and thus the fault and correct.

2 shows the automated self-referencing anomaly detection step according to the invention by means of an example of a value x ₀ which makes a sudden value jump in the course of a plurality of cycles after a certain cycle (in this case, for example, cycle 25). During the first 24 cycles, the value x, which may be, for example, a pressure value, a viscosity or another value of an injection molding process, that is generally a value of a process variable, is arranged within a value range of 20 to 21. These values x are assigned an average reference (dot-dash line) and a mean standard deviation (dashed line). From cycle 25, a value jump upwards into the range between 23 and 24 takes place, whereby in the further course from cycle 25 all values lie in this range.

Thus, the value jump from cycle 25 to cycle 26 represents an anomaly, which is not a singular anomaly but represents a persistent anomaly. Thus, it is not an outlier, but - as already mentioned above - by a jump in value, for example, if the value relates to a viscous sity, can be attributed to a change in the injection speed of the injection molding machine.

The self-referencing anomaly detection will be briefly explained below with reference to FIG. A batch production of identical parts (series parts) in the piece-based process, such as For example, in an injection molding machine, the characteristic that a process is stable only when the process characteristics for each cycle without T rend and without too large fluctuations. This property can be used to automatically detect all deviating events, eg. B. jumps, gradients, outliers, gradients, superimposed vibrations and the like to recognize and as anomalous to bewer th and mark or evaluate. For an automatic reference, the "natural" fluctuation is used, whereby each value has such a natural, in particular unavoidable, fluctuation due, for example, to slightly fluctuating environmental conditions or sensor noise. Such a natural fluctuation represents the best possible achievable stability of the process characteristic and is defined as such. As a measure of this, for example, the best stability achieved in the past can be used. This can be readily extrapolated to determine an assumed best achievable stability value for the future. However, this reference must necessarily be formed anew for certain events, for example when environmental conditions and / or other essential process parameters have changed. Such changes can z. B. a longer stand still the machine or a tool change or a material change or setting up the machine in other environmental conditions. By automating the monitoring and eliminating the manual limit setting, all values of the process can be monitored. The system is thus selftreferencing or self-learning and presents such anomalies in relation to a reference and independently decides on the use of a possibly new provisional reference compared to a previous current reference.

Reference to the figure 3 will now be explained briefly in connection with Figure 1 described Anoma case of an injection molding. In the injection molding process, for example, the plasticizing torque of the plasticizing screw can be measured with simple sensors. This represents a first value from a first data source (process size: "plasticizing torque"). Also, the melt pressure of the plastic melt can be easily measured at a variety of locations. The mass pressure is thus a further characteristic number or a further process variable. The tool wall temperature is also a measurable and in the present example a measured process variable in a simple manner. All measured or measurable process variables are shown in FIG. 3 in closed circular symbols. One knows now also that z. As a material viscosity, which is not so easy to measure in the concrete process, on the plasticizing as well as on the melt pressure effects, but not on the mold wall temperature.

If the anomaly detection modules of the individual parameters (process variables) "plasticizing torque" and "mass pressure" establish an abnormality for their values x ₀ ... X _j , the "tool wall temperature" remains inconspicuous, an existing computer system can automatically conclude that the im Model contained but not measured material (and / or its changed viscosity) must be the cause. This task falls to the root cause analysis and is filed there with appropriate rules. On the other hand, the parameters "cavity wall temperature" and "Melt pressure" anomalies and the "plasticizing torque" will not be the cause in the "mold wall temperature". Such an if-then relationship is also stored in the form of rules in the rules of the root cause analysis of the method according to the invention.

By means of a corresponding systematic processing, the user receives an indication of the cause in accordance with the invention and does not first have to analyze, weight and evaluate the often existing large number of individual anomalies in order to arrive at a corresponding cause result.

The inventive method can be readily z. B. be carried out via an interface to an injection molding machine, the interface characteristic values for each cycle to an external or internal computer system z. B. a Rechenein unit / controller on or in an injection molding machine sends. Such a computer system includes, for example, algorithms for evaluating various anomalies based on the automatically formed reference. The patterns of resulting anomalies are interpreted by the second algorithm and summarized into a diagnosis. This diagnosis is then the operator via the machine display or via a network / Internet z. B. mobile device such as a smartphone or tablet computer let and optionally displayed there. There you can z. B. also collected or received over a larger number of machines away and sorted, so that the inventive method, for example, simultane- ous occurrence of the same causes on a plurality of machines and a simplified cause analysis and cause research and repair for larger machine collections, eg , B. in a machine park, in a simple manner.

FIG. 4 schematically shows a flow chart of a reference generator. Such a reference generator according to the invention a variety, at least the most important process variables that are necessary for anomaly detection, zugeord net. The reference generator is supplied with a new value x _{0 on the} input side. On the basis of the new value x ₀ , new reference coefficients ri ^* ... R _n ^* belonging to this value x _{0 are} determined. Examples of such reference key figures can For example, as already mentioned, an arithmetic mean value or a standard deviation or further values that are preferably mathematically calculated from the values x ₀ can be.

At the same time, on the basis of the new value x _0, at least one valuation ratio bi ... bi is calculated. For example, a grading of the value curves of the values x ₀ ... X _j can be used as the evaluation parameter.

In the calculation of the new reference key figures ri ^* ... r _n ^* and in the calculation of the evaluation parameters bi ... b, of course, the last values Xi ... X _{j of} the process variable, which are opposite to the newly entered ones, flow in naturally and with particular preference Value x _{0 are} in the past.

Are charged with the new reference indicators ri ^* ... ^* r _n and the calculated evaluation parameter bi ... bi, a comparison with the current reference, which were formed from the or a plurality of earlier previous values Xi ... X _j, performed , Together with the comparison, an evaluation of additional criteria takes place, which can be done, for example, by means of the evaluation parameters bi ... b. Such an additional criterion may be, for example, the stability of the process. In the framework of the comparison of the calculated new reference numbers ri ^* ... _N ^* with the existing current reference n - r _n , the question is answered as to whether the new reference is better than the current (current) reference, in particular whether the new one Reference can better represent or represent the process or the course or the expected course of the corresponding process variable in the future than the current reference n ... r _n . If this is the case (yes), the current reference is n ... r _n _n ^* replaced by the new reference r ^* r ... so that the new reference r ^* r ... _n ^* the new current Refe rence ri ... becomes _n .

If this is not the case, the old reference, ie the old current reference n ... r _n , is retained.

The further process monitoring now takes place with the previous (current) reference n ... r _n or with the renewed updated reference ri ... r _n . The new value x _{0 of} the process variable in question is - as shown schematically in Figure 5 - together with the current, ie the current reference or the currently replaced reference n ... r _{n of} the Anomaliebewertung fed. The new value x ₀ is clearly marked as an anomaly by comparison with the corresponding applicable reference values n ... r _n and, where appropriate, taking into account past values Xi ... x _k in the context of the anomaly assessment or provided with a certain likelihood likelihood. Such an anomaly probability is assigned to the offending value x ₀ , if it is one, so that the value x _{0 is} either marked as anomaly or not (0 or 1 decision) or the value x _{0 of} the corresponding process variable with a certain anomaly truth probability (0 to 100%).

Claims

claims

1. A method for automatic process monitoring and / or process diagnosis of a piece-based process, in particular a manufacturing process, in particular a special injection molding process with the steps:

a) performing an automated reference finding to obtain reference values (h ... r _n ) from values (x ₀ ... Xj) of at least one process variable; b) performing anomaly detection on the basis of the reference values (n ... r _n ) found in step (a);

c) Carrying out an automated root cause analysis and / or an automated fault diagnosis on the basis of a qualitative model of process relationships and / or on the basis of dependencies of different process variables on each other.

2. The method according to claim 1,

characterized in that

a result of the root cause analysis and / or the fault diagnosis is output at an output device or a result of the root cause analysis / fault diagnosis is automatically processed further, eg. B. in a machine control and / or in a machine park control and / or in a controller for influencing a machine environment, for. B. a hall heating / air conditioning.

3. The method according to claim 1 or 2,

characterized in that

Step a) comprises at least one or more of the substeps listed below:

a1) Evaluation of process values x ₀ ... X _j of process variables over several process cycles with respect to their suitability for use as reference by the calculation of evaluation parameters bi ... b, and application of defined rules, where as evaluation parameters bi ... b For example, the change trend of the values x ₀ ... X _{j of} the process variables, and / or variations of the process variables are used or a2) reference values ... r _n are automatically used as reference for automatic process monitoring and / or automatic process diagnostics. B. reflect the "natural" noise or uncertainty of the process variable, which has any process variable due to environmental conditions and / or sensor noise, or

a) if the preliminary reference values r ^* i ... r ^* _n formed from the process values x ₀ ... X _j of process variables are better than the currently best found reference values ... r _n on the basis of criteria and rules new best reference values found r _n be established or ri ...

a4) the reference values ^ ... r _n from step a3) are used to automatically z. B. jumps, gradients, outliers as anomalies to recognize, to be evaluated and / or, where appropriate, to mark or

a5) wherein the automatic reference, that is, the reference values ^ ... r _{n is} formed necessarily new at predetermined events, such vorbe certain events, for example, a prolonged shutdown of the process leading machine or a tool change can be.

4. The method according to any one of claims 1 to 3,

characterized in that

Each process variable is assigned a reference generator, which is equipped with a Ini tial reference.

5. The method according to claim 4,

characterized in that

a reference from a plurality of reference values (ri ... r _n), wherein the Refe rence values (x _\ ... r _n) values properties of a curve of values (x ₀ ... Xj) of the process variable reflect such. B. the standard deviation and / or the medi of the value.

6. The method according to claim 4 or 5,

characterized in that During the process, the reference values (ri ... r _n ) are adapted to the course of the process, which is determined by measurement, whereby a window of j values of the process variable is considered here.

7. The method according to claim 6,

characterized in that

from the j values of the process variable (j) provisional reference values (ri ^* ... r _n ^* ) and evaluation numbers (bi ... b,) are formed.

8. The method according to claim 7,

characterized in that

the evaluation numbers (bi ... b,) derivations, eg. For example, the slope or curvature of the course of the j values of the process variable over time are.

9. The method according to any one of claims 7 or 8,

characterized in that

from the evaluation numbers (bi ... b,) of the current reference values (n ... r _n ) and the provisional reference values (n ^* ... r _n ^* ) is determined by means of predetermined rules whether the current reference values (ri .. r _n ) or, in the future, the provisional reference values (ri ^* ... r _n ^* ) are used as new current reference values (ri ... r _n ).

10. The method according to any one of the preceding claims,

characterized in that

for each process variable an anomaly detection is provided, which uses the current reference values (ri ... r _n ) and / or past values of the process variable (xi ... x _k ) to determine an exceptional value, ie an anomaly or in terms of to assess their likelihood.

1 1. Method according to one of the preceding claims,

characterized in that a value of a process variable (x ₀ ) having a predetermined distance from current reference values (ri ... r _n ), e.g. B. more than three reference standard deviations from the reference mean is removed, as "anomaly" is ge.

12. The method according to any one of the preceding claims,

characterized in that

as the qualitative model used in step c) of claim 1 qualitative model of an injection molding process is used in the relationships between tween the process variables and / or dependencies between the process variables are included.

13. Machine, in particular injection molding machine, which has a machine control and facilities for monitoring and / or measurement of process variables, wherein the machine is adapted and configured to perform the method according to one of claims 1 to 12.

14. Machine park, in particular injection molding machine park, which has a machine control and devices for monitoring and / or measurement of process variables, wherein the machine park is set up and designed to carry out the method according to one of claims 1 to 12.