The deviations in the compound quality parameters and reasons causing them are recognized online during the compound mixing. This information is used for real time tuning of the mixing control parameters to compensate the influence of the process disturbing factors.

Re-equipment to newest hardware and state of the art automation infrastructure does not automatically eliminate productivity and quality problems in daily production. The professionalism of process specialists, their experience and deep engagement into daily process still are the major components of the rubber business success. Any process responsible person knows that factors like the properties of raw materials, environmental conditions, process equipment status have great impact on the mixing process, its parameters and output quality. There are also numerous factors like abnormal events and non standard situations that need to be handled in real time to allow acceptable execution of production process. Widely used conventional way of process control does not react in fair scale to the mentioned conditions, events and threats. The usage of predefined process parameters for each step of mixing process (time, temperature, energy settings e.t.c.) does not provide possibility to compensate automatically influence of changes in properties of incoming materials, environmental conditions and the state of equipment on quality and productivity. The abnormal events handling cannot be provided as well except very simple cases.
In ideal case the process control system should execute functions of the experienced process engineer who is always available and who undertakes many duties simultaneously: stays near by the mixer all the time being fully aware what is going inside mixer and on mills, reacts in real time on changes in conditions by choosing proper control actions. He changes rotor speed, chooses right moments of materials introduction, corrects discharge temperatures, gives command to executes the movements of the ram in proper time. If to develop the dream further on, this process engineer should have abilities to know actual properties of currently processed batch quite before the sample is tested in laboratory.
Mixcont Software tools is an intelligent virtual assistant to process engineer (Intelligent Process Analyser) and its add-on module which is able to execute above mentioned control functions in real time (Predictive Mixing Process Control), which compensates for changes in raw materials properties, environment and machinery conditions by proper control actions, warns for output quality gives their detailed description and shows their roots.

Who will profit[2,3,6]
Process engineers and quality responsible personnel get tracing through the complete production process from raw materials to Laboratory; from the process mixing curves to test values, from out of specs lab values to the particular stage in process that caused problem. The impact of any production step that is normally not estimated (master batch mixing, remix, mills/extruder) is defined now and used for problem resolving analysis. Factory and production managers get cost efficient mixing by saving time and energy, reduced scrap rate and amount of remixed batches, essentially improved batch-to-batch uniformity and therefore quality of final goods. Transparency over the production factors by getting KPIs (Key Performance Indicators) and OEE (Overall Equipment Efficiency) in most complete form and exactly in time.Workers at Mixer/Mills get tracking on tendencies for running process from batch to batch and clear information in due time and about the problems or potential threats. The maintenance personal get alarms in proper time with precise focus on the problem occurred. Without installation of expensive equipment the rubber compounds producers can substantially enchance productivity and quality of the production.

Intelligent Analyzer – How it works [1-7,9,10].

Intelligent Analyzer is dedicated to investigate the Mixing Process and compound properties in real time.
The mixing process related information comes in digital form from the mixing equipment via data I/O interfaces (PLC/OPC drivers) (Fig. 1). The information is stored in SQL database and is used for data online process analyzing /reporting and also for predictive process control. The portion of information that describes the mixing process in total {(?-l(t)} of single batch is multiple array of process data in form of (i) Mixing Curves ?i(t): ram position, motor torque, rotor speed, temperatures from thermocouples, cooling temperatures, ram pressure and also numerous of digital signals are included.
There are few data processing stages that should be applied to get essential information about process behavior from mixing curves to be used in practice. The first one is check-in of integrity and correctness of the received information if the signals demonstrate correct logic, physical and feasible process behavior. The next one is to recognize the process steps online by using analog and digital signals from the process. The process is split into a number of logical stages J. The set of interim parameters are identified, in form of functional {?-ij (?-l(t)} of entire process from batch start to time moment (t), where (j) is the stage of mixing process[3,9]. It means, for example, that characteristics for the last mixing stage are dependent on behavior of the whole process before calculated moment – parameters of mastication stage, mixing after the oil and/or C/B introduction e.t.c.
These functional are calculated for time and frequency domains for the main informative parameters like torque, ram position, temperatures, ram pressure and their combinations. The aggregate quality parameters Qn(t) = Fn ({?-ij (?-(t))}) can be built as functions of named functional and in our model they related to process parameters of Dispersion/Distribution/Viscosity (DDV).
During the real time process control only differences in quality parameters ?Qnk(t)= Fnk({??-ijk(?-(t))}) as function of differences in process parameters are calculated from batch (k) to batch (k+1) (or to „model batch“ definition see below) are calculated, which simplifies the mathematical model of the process.
The time „t“ in above formulas is „effective mixing time“ – it is time when mixing forces (high enough to be considered) are applied to the batch. „Effective mixing time“ usually differs from batch to batch of the same recipe mixed with the same procedure. This is the coordinate which allows to compare two different batches. There is no surprise for any process engineer that, if the profiles of mixing process curves recorded during the mixing of 2 different batches are similar, the viscosity, dispersion and distribution of the batches are similar as well. At the same time the difference in the properties of the batches causes difference in mixing curves. The experienced process engineer, who uses the process curves in his daily practice, can rank diagrams (from chosen batches) of the mixing curves in order, for example, from the highest to the lowest dispersion. If dispersion of one of these batches is measured and can be characterized as „good enough“ (this batch can be considered as template batch), the process engineer can see which batches have dispersion not worse than „good enough“. Practically the process engineer can chose template batch by using test results information or getting feedback from final goods production. So, the evaluation of difference in the mixing curves is the possible way to evaluate the difference in properties between the batches of the same compound.
Intelligent Analyzer software is able to convert the difference in the Mixing curves, into the difference in process parameters including output quality indexes: dispersion, distribution , viscosity.
Due to the fact that MixCont Intelligent Analyzer incorporates the laboratory test data, it is easy to compare estimated values and laboratory test data. Analyzer scales internal Viscosity (Dispersion/Distribution) values to laboratory tests values scale. The analysis shows great level of correlation between MixCont estimation of „Viscosity Like“ values and Mooney Viscometer test values. „Mooney-like“ and Mooney test values for the same batch usually are not equal, but trends of Mooney-like and Mooney test values coincide if to compare enough number of batches (for instance 30). It was discovered that evaluation done by MixCont software can provide in some cases even more objective quality characteristics versus laboratory test values due to the fact that MixCont software analyses the whole batch, whereas the laboratory devices test few grams of rubber.

Predictive Process Control [1,2,6-9]
In fact Mixcont Predictive Control Module is Add-on to extend the Intelligent Analyzer and is used for implementation of real time process control. This module generates output control signals (settings for rotor speed, ram pressure, ram position, to add materials, discharge, discharge temperatures margines e.t.c.). The main task of the Module is to choose the optimal control strategy to reach the goal: to achieve requested quality parameters for minimum mixing time or/and energy for each batch. The optimal strategy for the each next moment is dependent on actual conditions and decisions taken at previous moments of time. So the Predictive Control system forecasts at current moment how the properties of the compound will progress (Fig. 2) , if different ways of mixing will be used during the rest part of the mixing time. These calculations are based on techniques which perform the set of all possible/ feasible strategies and can be applied for current and next moments of time. The goal of optimization is to choose the optimal alternative within the array of all feasible strategies. For choosing the optimal strategy MixCont Predictive Control uses Pareto optimization and/or dynamic programming techniques. The preparation of the complete set of possible and feasible control strategies (non dominated strategies ) for the particular moment is based on accumulated knowledge about process control, specific features of particular mixing line, mixing behavior of the particular compound, particular order and current batch.

Self learning abilities of control [4-6]
The mixing behavior of each compound is described by its own „model batch“, which is the accumulated knowledge about mixing behavior of the compound formalized in form of mixing curves and set of numeric parameters (see functional above). „Model batch“ is permanently updated following the noticeable changes in the mixing behavior of the compound caused by , for example, input polymer properties, outside temperatures, cooling conditions and many others. Quality goals are included into the set of „model“ parameters as well. The quality goals and initial set of parameters are defined by choosing template batch for the initial „model“, then during mixing the Adaptive Control System recalculates the „model “ set of parameters keeping the quality goals. In case of the „model“ parameters coincide to the last mixed batch the „model“ batch remains the same.

Adaptive control of the same compound in different outside conditions

This example shows how Mixcont Adaptive Control mixed the final batches of the same compound from two orders. The first of them was mixed when temperature outside was plus 20 °C and the other when around 0 °C. Difference in outside temperatures influenced the temperature of the master batches used as input material. The first master batch was much warmer and consequently much softer than the second one. MixCont Adaptive Control recognized the difference in the softness of the input material due to the difference in the mixing behavior of the batches (Fig.5) and reacted to it in the following way:
The rotors speed at the beginning of mixing of the harder (colder) batch was set higher in order to heat up and soften the batch quicker. As soon as temperature become high, the speed was decreased to keep the batch under high, but not dangerous temperature until the ram reaches the end position and the whole batch gets involved into the mixing. Then the speed was controlled with the aim not to allow the temperature to reach the discharge limit before homogeneity of the batch achieves acceptable level. The mixing time for the cold batch was considerably longer, because it took longer time to soften and compact the batch until all the materials were involved into the mixing.
The quality parameters of the both batches met quality requirements. The changes in process control were effected automatically without human participation. If the cold batch is mixed in the same way as the warm one, the cold batch would be discharged before the ram reached the end position, so homogeneity of the batch would be very poor.

Respond of Adaptive control to increase in polymer viscosity

Tyre recipe, master batch mixing , the natural rubber from a new delivery is used for the actual order. After 40 sec of the 1-st batch has being mixed in the mixer (moment t0) MixCont Analyzer recognized that if to mix the batch in the same way as the batches of the previous order, the viscosity of the batch will be considerably higher than for the „model“ batch (Fig. 4). Dispersion and distribution are expected to achieve the requested values much quicker than viscosity and there is high probability that polymer added into the batch has much higher viscosity. As soon as it is defined control system takes decision to increase discharge temperature for this batch, to add 2 ram sweeps and increase the rotor speed. It is done because higher rotor speed and higher viscosity of the batch provide more efficient mixing. This will speed up the process of viscosity decrease.
As the result viscosity of actual batch achieves the accepted level within the same time as the previous batches in spite of essentially higher viscosity of natural rubber. If this batch is mixed with the same mixing procedure as the previous batches, it would have abnormally high viscosity and should be remixed to make its output properties acceptable. The analysis of mixing behavior of the batch and corresponding recalculation for the control strategy is done each second. If the behavior of calculated parameters (viscosity, dispersion and others ) is in accordance to forecast made at previous second, the control system follows already chosen control strategy, if not, the strategy is recalculated and adjusted to the changed conditions.

Respond of Adaptive control to decrease in polymer viscosity

After a batch has been actively mixed for about 40 seconds, the Intelligent Analyzer defines that the polymer (natural rubber) in the actual batch is of lower viscosity than in the previous batches, so the system predicts that requested viscosity level will be achieved faster, but dispersion will improve slower in the softer batch and it will take longer time to get the requested dispersion than the requested viscosity. Thus, the dispersion is considered as critical parameter for this batch. As soon as it achieves acceptable level, the batch can be discharged. Therefore, the system decides (see moment t0) to arrange as favorable conditions as possible for the dispersion progress: it decreases the rotor speed ) to slow down the batch heating and softening at the beginning of mixing to use higher shear forces while the batch is colder and harder with the aim to speed up the dispersion progress. But after a while when the batch gets hotter and starts to soften actively anyway (time t*), the system increases the speed step by step to intensify the dispersion in the softer mixture. As the result the dispersion improves faster than it was predicted on 40 th second of active mixing (t0). As soon as the dispersion level comes close to the requested level, system steeply increases the rotor speed to accelerate the temperature growth in order to achieve minimum allowed limit of discharge temperature and requested dispersion level simultaneously (time moment t1).

Evaluation of quality properties online in details

A master batch has been just discharged from the mixer. Then mixing of the next batch is started. First high amount of polymer is added. During the mastication phase the absolute value of torque of the actual batch is higher (Fig. 4). Temperature grows faster in spite of rotor speed is a bit lower for this batch, the ram is on the end position in the both cases, the ram pressure is the same. Such features of the signals behavior can be the indication of three alternative cases: viscosity of the polymer in the actual batch is higher; the polymer is colder; higher amount of polymer is added to the mixer for this batch than for previous one. How to recognize the correct alternative?
If to consider one more feature – the torque fluctuations – the third alternative shall be excluded. Why ? The Amplitude-Frequency characteristics of the torque signal indicate the fact of harder polymer, but not of its higher volume. Due to the torque values show higher amplitude and lower frequency for the current batch in comparison to the previous one. This is feasible , the phenomena is well known to experienced process engineers. So, after 1 min of mixing the signal analysis concludes that only two alternatives are still possible: polymer added into actual batch is either colder or it is of higher viscosity. Which alternative is correct will be defined when the batch will get warmer and will start to soften. This fact will be indicated by torque decrease.
If the torque decreases actively and comes to the level of previous batches soon – this indicates that the polymer was colder at the moment it was put into the mixer. But, if torque decreases slower than during the mixing of the previous batches and stays over their level – the polymer added into the actual batch is of higher viscosity. The higher difference in absolute values of torque and the rate of torque decrease, the higher is the difference in polymer viscosity between the batches.

Evaluation of difference in dispersion between two bathes

The harder the batch, the faster dispersion progresses during the mixing. The opposite statement is also correct. Harder batch require higher torque from the electric drive to mix it , higher torque which is applied to the batch causes higher shear forces and consequently causes more active process of destruction of carbon black agglomerates. The lower the torque , the longer the batch shall be mixed to achieve the same level. In order to define how the dispersion level of the actual batch differs from the any other Control Program compares the torque during the mixing and mixing time for the both batches. The dispersion evaluation approach above was tested for many different compounds at different factories. These tests have proved that dispersion values estimated by Mixcont system have very good correlation with the results provided by correspondent test devices, like disperGrader and/or laboratory results for tensile strength.


The MixCont Intelligent Analyzer and Predictive Process Control systems are build based on the integral process control experience and engineers common sense by using disciplines like mechanics, statistics, pattern recognition, signal processing and optimization models.
The MixCont technologies are efficient tools bringing unique benefits for users:

  • Decrease production time and energy per produced unit (5-20?%) for the mixing line
  • Enhanced batch -to-batch uniformity essentially improves quality and productivity of final goods,
  • Definition and display batch process characteristics online: Viscosity/Dispersion/Distribution before laboratory tests done, advice correction of fill factor online
  • Adjustment process control parameters in real time, if current mixing conditions have changed, like polymer properties, equipment performance
  • Automatically define the process related problems and abnormal events if occur with clear visualization of the roots.
  • Calculation of overall equipment efficiency and key performance factors for problem resolving and debottlenecking by using process mixing curves , therefore with higher level of precision than traditional ways
  • Mixcont with its partner offers turn key solution for mixing line which embraces complete automation for mixing line including supply and turn key installation of software and hardware. All components for mixing line are included: weighing/dosing/mixer /mills/batch -off, centralised SQL database server for collecting and integration of information with inbuilt Intelligent Process Analyser and Predictive Process control modules . The Laboratory Management system, Recipe/Materials manipulation, warehouse automation and ERP interfaces are extra options and available as components for turn-key supply.

About the author

Dr. S. Brassas, M. Sarbatova, Mixcont,Trelleborg, Sweden